🔭Data Science: Noise (Just the Quotes)

"Information that is only partially structured (and therefore contains some 'noise' is fuzzy, inconsistent, and indistinct. Such imperfect information may be regarded as having merit only if it represents an intermediate step in structuring the information into a final meaningful form. If the partially Structured information remains in fuzzy form, it will create a state of dissatisfaction in the mind of the originator and certainly in the mind of the recipient. The natural desire is to continue structuring until clarity, simplicity, precision, and definitiveness are obtained." (Cecil H Meyers, "Handbook of Basic Graphs: A modern approach", 1970)

"To understand the need for structuring information, we should examine its opposite - nonstructured information. Nonstructured information may be thought of as exists and can be heard (or sensed with audio devices), but the mind attaches no rational meaning to the sound. In another sense, noise can be equated to writing a group of letters, numbers, and other symbols on a page without any design or key to their meaning. In such a situation, there is nothing the mind can grasp. Nonstructured information can be classified as useless, unless meaning exists somewhere in the jumble and a key can be found to unlock its hidden significance." (Cecil H Meyers, "Handbook of Basic Graphs: A modern approach", 1970)

"Experiments usually are looking for 'signals' of truth, and the search is always ham pered by 'noise' of one kind or another. In judging someone else's experimental results it's important to find out whether they represent a true signal or whether they are just so much noise." (Robert Hooke, "How to Tell the Liars from the Statisticians", 1983)

"Neither noise nor information is predictable." (Ray Kurzweil, "The Age of Spiritual Machines: When Computers Exceed Human Intelligence", 1999)

"Data are collected as a basis for action. Yet before anyone can use data as a basis for action the data have to be interpreted. The proper interpretation of data will require that the data be presented in context, and that the analysis technique used will filter out the noise."  (Donald J Wheeler, "Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"Data are generally collected as a basis for action. However, unless potential signals are separated from probable noise, the actions taken may be totally inconsistent with the data. Thus, the proper use of data requires that you have simple and effective methods of analysis which will properly separate potential signals from probable noise." (Donald J Wheeler, "Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"No matter what the data, and no matter how the values are arranged and presented, you must always use some method of analysis to come up with an interpretation of the data. While every data set contains noise, some data sets may contain signals. Therefore, before you can detect a signal within any given data set, you must first filter out the noise." (Donald J Wheeler," Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"We analyze numbers in order to know when a change has occurred in our processes or systems. We want to know about such changes in a timely manner so that we can respond appropriately. While this sounds rather straightforward, there is a complication - the numbers can change even when our process does not. So, in our analysis of numbers, we need to have a way to distinguish those changes in the numbers that represent changes in our process from those that are essentially noise." (Donald J Wheeler, "Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"While all data contain noise, some data contain signals. Before you can detect a signal, you must filter out the noise." (Donald J Wheeler, "Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"Prediction is rarely perfect. There are usually many unmeasured variables whose effect is referred to as 'noise'. But the extent to which the model box emulates nature's box is a measure of how well our model can reproduce the natural phenomenon producing the data." (Leo Breiman, "Statistical Modeling: The Two Cultures", Statistical Science 16(3), 2001)

"The acquisition of information is a flow from noise to order - a process converting entropy to redundancy. During this process, the amount of information decreases but is compensated by constant re-coding. In the recoding the amount of information per unit increases by means of a new symbol which represents the total amount of the old. The maturing thus implies information condensation. Simultaneously, the redundance decreases, which render the information more difficult to interpret." (Lars Skyttner, "General Systems Theory: Ideas and Applications", 2001)

"In fact, an information theory that leaves out the issue of noise turns out to have no content." (Hans Christian von Baeyer, "Information, The New Language of Science", 2003)

"This phenomenon, common to chaos theory, is also known as sensitive dependence on initial conditions. Just a small change in the initial conditions can drastically change the long-term behavior of a system. Such a small amount of difference in a measurement might be considered experimental noise, background noise, or an inaccuracy of the equipment." (Greg Rae, Chaos Theory: A Brief Introduction, 2006)

"[…] noise is the random variation that results from measurement error or other causes not accounted for. It is always present in a time series to some degree, although we cannot observe it directly." (Galit Shmueli, "Practical Time Series Forecasting: A Hands-On Guide", 2011)

"Some forecasting methods directly model these components by making assumptions about their structure. For example, a popular assumption about trend is that it is linear or exponential over parts, or all, of the given time period. Another common assumption is about the noise structure: many statistical methods assume that the noise follows a normal distribution. The advantage of methods that rely on such assumptions is that when the assumptions are reasonably met, the resulting forecasts will be more robust and the models more understandable." (Galit Shmueli, "Practical Time Series Forecasting: A Hands-On Guide", 2011)

"Overfitting means that the model is not only fitting the systematic component of the data, but also the noise. An over-fitted model is therefore likely to perform poorly on new data." (Galit Shmueli, "Practical Time Series Forecasting: A Hands-On Guide", 2011)

"Data analysis is not generally thought of as being simple or easy, but it can be. The first step is to understand that the purpose of data analysis is to separate any signals that may be contained within the data from the noise in the data. Once you have filtered out the noise, anything left over will be your potential signals. The rest is just details." (Donald J Wheeler," Myths About Data Analysis", International Lean & Six Sigma Conference, 2012)

"Distinguishing the signal from the noise requires both scientific knowledge and self-knowledge." (Nate Silver, "The Signal and the Noise: Why So Many Predictions Fail-but Some Don't", 2012)

"Economists should study financial markets as they actually operate, not as they assume them to operate - observing the way in which information is actually processed, observing the serial correlations, bonanzas, and sudden stops, not assuming these away as noise around the edges of efficient and rational markets." (Adair Turner, "Economics after the Crisis: Objectives and means", 2012)

"Finding patterns is easy in any kind of data-rich environment; that's what mediocre gamblers do. The key is in determining whether the patterns represent signal or noise." (Nate Silver, "The Signal and the Noise: Why So Many Predictions Fail-but Some Don't", 2012)

"The signal is the truth. The noise is what distracts us from the truth." (Nate Silver, "The Signal and the Noise: Why So Many Predictions Fail-but Some Don't", 2012)

"Typically, most outlier detection algorithms use some quantified measure of the outlierness of a data point, such as the sparsity of the underlying region, nearest neighbor based distance, or the fit to the underlying data distribution. Every data point lies on a continuous spectrum from normal data to noise, and finally to anomalies [...] The separation of the different regions of this spectrum is often not precisely defined, and is chosen on an ad-hoc basis according to application-specific criteria. Furthermore, the separation between noise and anomalies is not pure, and many data points created by a noisy generative process may be deviant enough to be interpreted as anomalies on the basis of the outlier score. Thus, anomalies will typically have a much higher outlier score than noise, but this is not a distinguishing factor between the two as a matter of definition. Rather, it is the interest of the analyst, which regulates the distinction between noise and an anomaly." (Charu C Aggarwal, "Outlier Analysis", 2013)

"A complete data analysis will involve the following steps: (i) Finding a good model to fit the signal based on the data. (ii) Finding a good model to fit the noise, based on the residuals from the model. (iii) Adjusting variances, test statistics, confidence intervals, and predictions, based on the model for the noise." (DeWayne R Derryberry, "Basic data analysis for time series with R", 2014)

 "The random element in most data analysis is assumed to be white noise - normal errors independent of each other. In a time series, the errors are often linked so that independence cannot be assumed (the last examples). Modeling the nature of this dependence is the key to time series." (DeWayne R Derryberry, "Basic data analysis for time series with R", 2014)

"There are many roads to statistical significance; if data are gathered with no preconceptions at all, statistical significance can obviously be obtained even from pure noise by the simple means of repeatedly performing comparisons, excluding data in different ways, examining different interactions, controlling for different predictors, and so forth. Realistically, though, a researcher will come into a study with strong substantive hypotheses, to the extent that, for any given data set, the appropriate analysis can seem evidently clear. But even if the chosen data analysis is a deterministic function of the observed data, this does not eliminate the problem posed by multiple comparisons." (Andrew Gelman & Eric Loken, "The Statistical Crisis in Science", American Scientist Vol. 102(6), 2014)

"A predictive model overfits the training set when at least some of the predictions it returns are based on spurious patterns present in the training data used to induce the model. Overfitting happens for a number of reasons, including sampling variance and noise in the training set. The problem of overfitting can affect any machine learning algorithm; however, the fact that decision tree induction algorithms work by recursively splitting the training data means that they have a natural tendency to segregate noisy instances and to create leaf nodes around these instances. Consequently, decision trees overfit by splitting the data on irrelevant features that only appear relevant due to noise or sampling variance in the training data. The likelihood of overfitting occurring increases as a tree gets deeper because the resulting predictions are based on smaller and smaller subsets as the dataset is partitioned after each feature test in the path." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015

"A signal is a useful message that resides in data. Data that isn’t useful is noise. […] When data is expressed visually, noise can exist not only as data that doesn’t inform but also as meaningless non-data elements of the display (e.g. irrelevant attributes, such as a third dimension of depth in bars, color variation that has no significance, and artificial light and shadow effects)." (Stephen Few, "Signal: Understanding What Matters in a World of Noise", 2015)

"Data contain descriptions. Some are true, some are not. Some are useful, most are not. Skillful use of data requires that we learn to pick out the pieces that are true and useful. [...] To find signals in data, we must learn to reduce the noise - not just the noise that resides in the data, but also the noise that resides in us. It is nearly impossible for noisy minds to perceive anything but noise in data." (Stephen Few, "Signal: Understanding What Matters in a World of Noise", 2015)

"When we find data quality issues due to valid data during data exploration, we should note these issues in a data quality plan for potential handling later in the project. The most common issues in this regard are missing values and outliers, which are both examples of noise in the data." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, worked examples, and case studies", 2015)

"Information theory leads to the quantification of the information content of the source, as denoted by entropy, the characterization of the information-bearing capacity of the communication channel, as related to its noise characteristics, and consequently the establishment of the relationship between the information content of the source and the capacity of the channel. In short, information theory provides a quantitative measure of the information contained in message signals and help determine the capacity of a communication system to transfer this information from source to sink over a noisy channel in a reliable fashion." (Ali Grami, "Information Theory", 2016)

"High-bias models typically produce simpler models that do not overfit and in those cases the danger is that of underfitting. Models with low-bias are typically more complex and that complexity enables us to represent the training data in a more accurate way. The danger here is that the flexibility provided by higher complexity may end up representing not only a relationship in the data but also the noise. Another way of portraying the bias-variance trade-off is in terms of complexity v simplicity." (Jesús Rogel-Salazar, "Data Science and Analytics with Python", 2017)

"Repeated observations of the same phenomenon do not always produce the same results, due to random noise or error. Sampling errors result when our observations capture unrepresentative circumstances, like measuring rush hour traffic on weekends as well as during the work week. Measurement errors reflect the limits of precision inherent in any sensing device. The notion of signal to noise ratio captures the degree to which a series of observations reflects a quantity of interest as opposed to data variance. As data scientists, we care about changes in the signal instead of the noise, and such variance often makes this problem surprisingly difficult." (Steven S Skiena, "The Data Science Design Manual", 2017)

"The dichotomy of mathematical vs. statistical modeling says more about the culture of modeling and how different disciplines go about thinking about models than about how we should actually model ecological systems. A mathematician is more likely to produce a deterministic, dynamic process model without thinking very much about noise and uncertainty (e.g. the ordinary differential equations that make up the Lotka-Volterra predator prey model). A statistician, on the other hand, is more likely to produce a stochastic but static model, that treats noise and uncertainty carefully but focuses more on static patterns than on the dynamic processes that produce them (e.g. linear regression)." (Ben Bolker, "Ecological Models and Data in R", 2007)

"Using noise (the uncorrelated variables) to fit noise (the residual left from a simple model on the genuinely correlated variables) is asking for trouble." (Steven S Skiena, "The Data Science Design Manual", 2017)

"The high generalization error in a neural network may be caused by several reasons. First, the data itself might have a lot of noise, in which case there is little one can do in order to improve accuracy. Second, neural networks are hard to train, and the large error might be caused by the poor convergence behavior of the algorithm. The error might also be caused by high bias, which is referred to as underfitting. Finally, overfitting (i.e., high variance) may cause a large part of the generalization error. In most cases, the error is a combination of more than one of these different factors." (Charu C Aggarwal, "Neural Networks and Deep Learning: A Textbook", 2018)

"[...] in the statistical world, what we see and measure around us can be considered as the sum of a systematic mathematical idealized form plus some random contribution that cannot yet be explained. This is the classic idea of the signal and the noise." (David Spiegelhalter, "The Art of Statistics: Learning from Data", 2019)

"We thus echo the classical Bayesian literature in concluding that ‘noninformative prior information’ is a contradiction in terms. The flat prior carries information just like any other; it represents the assumption that the effect is likely to be large. This is often not true. Indeed, the signal-to-noise ratios is often very low and then it is necessary to shrink the unbiased estimate. Failure to do so by inappropriately using the flat prior causes overestimation of effects and subsequent failure to replicate them." (Erik van Zwet & Andrew Gelman, "A proposal for informative default priors scaled by the standard error of estimates", The American Statistician 76, 2022)

"Visualizations can remove the background noise from enormous sets of data so that only the most important points stand out to the intended audience. This is particularly important in the era of big data. The more data there is, the more chance for noise and outliers to interfere with the core concepts of the data set." (Kate Strachnyi, "ColorWise: A Data Storyteller’s Guide to the Intentional Use of Color", 2023)

"Overfitting in machine learning is a condition in which a model learns the training data so well that it performs poorly on unseen or test data. The model essentially memorizes the noise or random fluctuations in the training data and fails to generalize its learning to new data. In terms of LLMs, overfitting could occur if the model excessively adjusts to the specifics of the training data, thereby losing its ability to generate sensible responses for unseen prompts. This could lead to the model generating too specific or narrowly tailored responses that do not correctly address the new prompts." (Sinan Ozdemir, "Quick Start Guide to Large Language Models: Strategies and Best Practices for Using ChatGPT and Other LLMs", 2024)

"Besides allowing us to compress a sparse representation into a denser vector, autoencoders are also widely used to process a signal in order to filter out noise and extract only a relevant (characteristic) signal. This can be very useful in many applications, especially when identifying anomalies and outliers." (Aldo Marzullo et al, "Graph Machine Learning" 2nd Ed., 2025)

🔭Data Science: Unsupervised Learning/Cluster Analysis (Just the Quotes)

"Multivariate techniques often summarize or classify many variables to only a few groups or factors (e.g., cluster analysis or multi-dimensional scaling). Parallel coordinate plots can help to investigate the influence of a single variable or a group of variables on the result of a multivariate procedure. Plotting the input variables in a parallel coordinate plot and selecting the features of interest of the multivariate procedure will show the influence of different input variables." (Martin Theus & Simon Urbanek, "Interactive Graphics for Data Analysis: Principles and Examples", 2009)

"Supervised learning, or function approximation, is simply fitting data to a function of any variety.  […] Unsupervised learning involves figuring out what makes the data special. […] Reinforcement learning involves figuring out how to play a multistage game with rewards and payoffs. Think of it as the algorithms that optimize the life of something." (Matthew Kirk,Thoughtful Machine Learning", 2015)

"A major challenge in unsupervised learning is evaluating whether the algorithm learned something useful. Unsupervised learning algorithms are usually applied to data that does not contain any label information, so we don’t know what the right output should be. Therefore, it is very hard to say whether a model 'did well'. [...] As a consequence, unsupervised algorithms are used often in an exploratory setting, when a data scientist wants to understand the data better, rather than as part of a larger automatic system. Another common application for unsupervised algorithms is as a preprocessing step for supervised algorithms. Learning a new representation of the data can sometimes improve the accuracy of supervised algorithms, or can lead to reduced memory and time consumption." (Andreas C Müller & Sarah Guido, "Introduction to Machine Learning with Python: A Guide for Data Scientists", 2017)

"Unsupervised transformations of a dataset are algorithms that create a new representation of the data which might be easier for humans or other machine learning algorithms to understand compared to the original representation of the data. A common application of unsupervised transformations is dimensionality reduction, which takes a high-dimensional representation of the data, consisting of many features, and finds a new way to represent this data that summarizes the essential characteristics with fewer features. A common application for dimensionality reduction is reduction to two dimensions for visualization purposes." (Andreas C Müller & Sarah Guido, "Introduction to Machine Learning with Python: A Guide for Data Scientists", 2017)

"A form of machine learning in which the goal is to identify regularities in the data. These regularities may include clusters of similar instances within the data or regularities between attributes. In contrast to supervised learning, in unsupervised learning no target attribute is defined in the data set." (John D Kelleher & Brendan Tierney,Data science", 2018)

"Unsupervised learning or clustering is a way of discovering hidden structures in unlabeled data. Clustering algorithms aim to discover latent patterns in unlabeled data using features to organize instances into meaningfully dissimilar groups." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"Machine Learning (ML) is a subfield of AI that enables computers to acquire knowledge and improve performance without explicit programming. Different from traditional programming, where you must write the code line by line to provide instructions, ML allows machines to learn from data and make predictions or take actions based on that learned information. This approach empowers machines to discover patterns, relationships, and insights that may not be immediately apparent to human programmers. ML relies on two types of learning: supervised learning and unsupervised learning." (Angelica Lo Duca, "Data Storytelling with Generative AI: Using Python and Altair", 2024) 

"Generally speaking, all the unsupervised embedding algorithms based on matrix factorization use the same principle. They all factorize an input graph expressed as a matrix in different components. The main difference between each method lies in the loss function used during the optimization process and the constraints/formulations posed for the V and H matrices. Indeed, different loss functions allow the creation of an embedding space that emphasizes specific properties of the input graph." (Aldo Marzullo et al, "Graph Machine Learning" 2nd Ed., 2025)

"In a nutshell, training LLMs involves optimizing a large number of parameters on very large datasets. This process, known as pretraining, typically employs unsupervised learning objectives, such as predicting missing words in a sentence (masked language modeling) or forecasting subsequent words (causal language modeling). As a side effect, the pre-training phase lets the model learn and 'understand' a language, resulting in a remarkable ability to generalize across various tasks, often achieving state-of-the-art performance. LLMs have demonstrated proficiency in a diverse array of applications, reflecting their versatility and depth of language understanding. Key areas include text generation, language translation, question answering, and summarization, among many others." (Aldo Marzullo et al, "Graph Machine Learning" 2nd Ed., 2025)

"The main difference between unsupervised and supervised embedding methods essentially lies in the task they attempt to solve. Indeed, if unsupervised shallow embedding algorithms try to learn a good graph, node, or edge representation in order to understand the underlying structure, the supervised algorithms try to find the best solution for a prediction task such as node classification, label prediction, or graph classification." (Aldo Marzullo et al, "Graph Machine Learning" 2nd Ed., 2025)

"Unsupervised classification, or cluster analysis, organizes observations into similar groups. Clusteranalysis is a commonly used, appealing, and conceptually intuitive statistical method. Some of its uses include market segmentation, where customers are grouped into clusters with similar attributes for targeted marketing; gene expression analysis, where genes with similar expression patterns are grouped together; and the creation of taxonomies for animals, insects, or plants. Clustering can be used as a way of reducing a massive amount of data because observations within a cluster can be summarised by its centre. Also, clustering effectively subsets the data thus simplifying analysis because observations in each cluster can be analysed separately." (Dianne Cook & Ursula Laa, "Interactively Exploring High-Dimensional Data and Models in R", 2026)

🔭Data Science: Semantic Networks (Just the Quotes)

"In comparison with Predicate Calculus encoding s of factual knowledge, semantic nets seem more natural and understandable. This is due to the one-to-one correspondence between nodes and the concepts they denote, to the clustering about a particular node of propositions about a particular thing, and to the visual immediacy of 'interrelationships' between concepts, i.e., their connections via sequences of propositional links." (Lenhart K Schubert, "Extending the Expressive Power of Semantic Networks", Artificial Intelligence 7, 1976)

"[…] semantic nets [are defined] as graphical analogues of data structures representing 'facts' in a computer system for understanding natural language." (Lenhart K Schubert," "Extending the Expressive Power of Semantic Networks", Artificial Intelligence 7, 1976)

"The advantage of semantic networks over standard logic is that some selected set of the possible inferences can be made in a specialized and efficient way. If these correspond to the inferences that people make naturally, then the system will be able to do a more natural sort of reasoning than can be easily achieved using formal logical deduction." (Avron Barr, Natural Language Understanding, AI Magazine Vol. 1 (1), 1980)

"We define a semantic network as 'the collection of all the relationships that concepts have to other concepts, to percepts, to procedures, and to motor mechanisms' of the knowledge"." (John F Sowa, Conceptual Structures, 1984)

"[…] semantic nets fail to be distinctive in the way they (1) represent propositions, (2) cluster information for access, (3) handle property inheritance, and (4) handle general inference; in other words, they lack distinctive representational properties (i.e., 1) and distinctive computational properties (i.e., 2-4). Certain propagation mechanisms, notably 'spreading activation', 'intersection search', or 'inference propagation' have sometimes been regarded as earmarks of semantic nets, but since most extant semantic nets lack such mechanisms, they cannot be considered criterial in current usage." (Lenhart K Schubert, "Semantic Nets are in the Eye of the Beholder", 1990)

"[…] the representational and computational strategies employed in semantic net systems are abstractly equivalent to those employed in virtually all state-of-the-art systems incorporating a substantial propositional knowledge base, whether they are described as logic-based, frame-based, rule-based, or some-thing else." (Lenhart K Schubert, "Semantic Nets are in the Eye of the Beholder", 1990)

"A semantic network or net represents knowledge as a net-like graph. An idea, event, situation or object almost always has a composite structure; this is represented in a semantic network by a corresponding structure of nodes (drawn as circles or boxes) representing conceptual units, and directed links (drawn as arrows between the nodes) representing the relations between the units. […] An abstract (graph-theoretic) network can be diagrammed, defined mathematically, programmed in a computer, or hard-wired electronically. It becomes semantic when you assign a meaning to each node and link. Unlike specialized networks and diagrams, semantic networks aim to represent any kind of knowledge which can be described in natural language. A semantic network system includes not only the explicitly stored net structure but also methods for automatically deriving from that a much larger structure or body of implied knowledge." (Fritz Lehman, "Semantic Networks",  Computers & Mathematics with Applications Vol. 23 (2-5), 1992)

"The essential idea of semantic networks is that the graph-theoretic structure of relations and. abstractions can be used for inference as well as understanding. […] A semantic network is a discrete structure as is any linguistic description. Representation of the continuous 'outside world' with such a structure is necessarily incomplete, and requires decisions as to which information is kept and which is lost." (Fritz Lehman, "Semantic Networks",  Computers & Mathematics with Applications Vol. 23 (2-5), 1992)

"Semantic networks are used to illustrate how people organize information in their memories. Such representations have been used by cognitive psychologists to understand and theorize how one retrieves and processes information from long-term memory. In AI, semantic networks can also be used as a knowledge representation scheme that programs can use to retrieve information efficiently just like humans do." (Robbie T Nakatsu, "Diagrammatic Reasoning in AI", 2010)

"Graphs can embed complex semantic representations in a compact form. As such, modeling data as networks of related entities is a powerful mechanism for analytics, both for visual analyses and machine learning. Part of this power comes from performance advantages of using a graph data structure, and the other part comes from an inherent human ability to intuitively interact with small networks." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"In Exploiting semantic knowledge graphs can support interpretability and explainability of nearly all AI model types (including DL models) by discovering and depicting semantic and non-obvious relationships or depicting an ML model in a simplified and more readable, explainable way., a Data Mesh solution organizes data around business domain owners and transforms relevant data assets (data sources) to data products that can be consumed by distributed business users from various business domains or functions. These data products are created, governed, and used in an autonomous, decentralized, and self-service manner. Self-service capabilities, which we have already referenced as a Data Fabric capability, enable business organizations to entertain a data marketplace with shopping-for-data characteristics." (Eberhard Hechler et al, "Data Fabric and Data Mesh Approaches with AI", 2023)

See also the Definitions

🔭Data Science: Missing Data (Just the Quotes)

"Place little faith in an average or a graph or a trend when those important figures are missing."  (Darell Huff, "How to Lie with Statistics", 1954)

"Missing data values pose a particularly sticky problem for symbols. For instance, if the ray corresponding to a missing value is simply left off of a star symbol, the result will be almost indistinguishable from a minimum (i.e., an extreme) value. It may be better either (i) to impute a value, perhaps a median for that variable, or a fitted value from some regression on other variables, (ii) to indicate that the value is missing, possibly with a dashed line, or (iii) not to draw the symbol for a particular observation if any value is missing." (John M Chambers et al, "Graphical Methods for Data Analysis", 1983)

"The progress of science requires more than new data; it needs novel frameworks and contexts. And where do these fundamentally new views of the world arise? They are not simply discovered by pure observation; they require new modes of thought. And where can we find them, if old modes do not even include the right metaphors? The nature of true genius must lie in the elusive capacity to construct these new modes from apparent darkness. The basic chanciness and unpredictability of science must also reside in the inherent difficulty of such a task." (Stephen J Gould, "The Flamingo's Smile: Reflections in Natural History", 1985)

"We often think, naïvely, that missing data are the primary impediments to intellectual progress - just find the right facts and all problems will dissipate. But barriers are often deeper and more abstract in thought. We must have access to the right metaphor, not only to the requisite information. Revolutionary thinkers are not, primarily, gatherers of facts, but weavers of new intellectual structures." (Stephen J Gould, "The Flamingo's Smile: Reflections in Natural History", 1985)

"[...] as the planning process proceeds to a specific financial or marketing state, it is usually discovered that a considerable body of 'numbers' is missing, but needed numbers for which there has been no regular system of collection and reporting; numbers that must be collected outside the firm in some cases. This serendipity usually pays off in a much better management information system in the form of reports which will be collected and reviewed routinely." (William H. Franklin Jr., Financial Strategies, 1987)

"We have found that some of the hardest errors to detect by traditional methods are unsuspected gaps in the data collection (we usually discovered them serendipitously in the course of graphical checking)." (Peter Huber, "Huge data sets", Compstat ’94: Proceedings, 1994)

"Unfortunately, just collecting the data in one place and making it easily available isn’t enough. When operational data from transactions is loaded into the data warehouse, it often contains missing or inaccurate data. How good or bad the data is a function of the amount of input checking done in the application that generates the transaction. Unfortunately, many deployed applications are less than stellar when it comes to validating the inputs. To overcome this problem, the operational data must go through a 'cleansing' process, which takes care of missing or out-of-range values. If this cleansing step is not done before the data is loaded into the data warehouse, it will have to be performed repeatedly whenever that data is used in a data mining operation." (Joseph P Bigus,"Data Mining with Neural Networks: Solving business problems from application development to decision support", 1996)

"If you have only a small proportion of cases with missing data, you can simply throw out those cases for purposes of estimation; if you want to make predictions for cases with missing inputs, you don’t have the option of throwing those cases out." (Warren S Sarle, "Prediction with missing inputs", 1998)

"Every statistical analysis is an interpretation of the data, and missingness affects the interpretation. The challenge is that when the reasons for the missingness cannot be determined there is basically no way to make appropriate statistical adjustments. Sensitivity analyses are designed to model and explore a reasonable range of explanations in order to assess the robustness of the results." (Gerald van Belle, "Statistical Rules of Thumb", 2002)

"The best rule is: Don't have any missing data, Unfortunately, that is unrealistic. Therefore, plan for missing data and develop strategies to account for them. Do this before starting the study. The strategy should state explicitly how the type of missingness will be examined, how it will be handled, and how the sensitivity of the results to the missing data will be assessed." (Gerald van Belle, "Statistical Rules of Thumb", 2002)

"Statistics depend on collecting information. If questions go unasked, or if they are asked in ways that limit responses, or if measures count some cases but exclude others, information goes ungathered, and missing numbers result. Nevertheless, choices regarding which data to collect and how to go about collecting the information are inevitable." (Joel Best, "More Damned Lies and Statistics: How numbers confuse public issues", 2004)

"A sin of omission – leaving something out – is a strong one and not always recognized; itʼs hard to ask for something you donʼt know is missing. When looking into the data, even before it is graphed and charted, there is potential for abuse. Simply not having all the data or the correct data before telling your story can cause problems and unhappy endings." (Brian Suda, "A Practical Guide to Designing with Data", 2010)

"Having NUMBERSENSE means: (•) Not taking published data at face value; (•) Knowing which questions to ask; (•) Having a nose for doctored statistics. [...] NUMBERSENSE is that bit of skepticism, urge to probe, and desire to verify. It’s having the truffle hog’s nose to hunt the delicacies. Developing NUMBERSENSE takes training and patience. It is essential to know a few basic statistical concepts. Understanding the nature of means, medians, and percentile ranks is important. Breaking down ratios into components facilitates clear thinking. Ratios can also be interpreted as weighted averages, with those weights arranged by rules of inclusion and exclusion. Missing data must be carefully vetted, especially when they are substituted with statistical estimates. Blatant fraud, while difficult to detect, is often exposed by inconsistency." (Kaiser Fung, "Numbersense: How To Use Big Data To Your Advantage", 2013)

"Quality without science and research is absurd. You can't make inferences that something works when you have 60 percent missing data." (Peter Pronovost, "Safe Patients, Smart Hospitals", 2010)

"The only thing we know for sure about a missing data point is that it is not there, and there is nothing that the magic of statistics can do change that. The best that can be managed is to estimate the extent to which missing data have influenced the inferences we wish to draw." (Howard Wainer, "14 Conversations About Three Things", Journal of Educational and Behavioral Statistics Vol. 35(1, 2010)

"There are several key issues in the field of statistics that impact our analyses once data have been imported into a software program. These data issues are commonly referred to as the measurement scale of variables, restriction in the range of data, missing data values, outliers, linearity, and nonnormality." (Randall E Schumacker & Richard G Lomax, "A Beginner’s Guide to Structural Equation Modeling" 3rd Ed., 2010)

"Missing data is the blind spot of statisticians. If they are not paying full attention, they lose track of these little details. Even when they notice, many unwittingly sway things our way. Most ranking systems ignore missing values." (Kaiser Fung, "Numbersense: How To Use Big Data To Your Advantage", 2013)

"Accuracy and coherence are related concepts pertaining to data quality. Accuracy refers to the comprehensiveness or extent of missing data, performance of error edits, and other quality assurance strategies. Coherence is the degree to which data - item value and meaning are consistent over time and are comparable to similar variables from other routinely used data sources." (Aileen Rothbard, "Quality Issues in the Use of Administrative Data Records", 2015)

"How good the data quality is can be looked at both subjectively and objectively. The subjective component is based on the experience and needs of the stakeholders and can differ by who is being asked to judge it. For example, the data managers may see the data quality as excellent, but consumers may disagree. One way to assess it is to construct a survey for stakeholders and ask them about their perception of the data via a questionnaire. The other component of data quality is objective. Measuring the percentage of missing data elements, the degree of consistency between records, how quickly data can be retrieved on request, and the percentage of incorrect matches on identifiers (same identifier, different social security number, gender, date of birth) are some examples." (Aileen Rothbard, "Quality Issues in the Use of Administrative Data Records", 2015)

"When we find data quality issues due to valid data during data exploration, we should note these issues in a data quality plan for potential handling later in the project. The most common issues in this regard are missing values and outliers, which are both examples of noise in the data." (John D. Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, worked examples, and case studies", 2015)

"There are other problems with Big Data. In any large data set, there are bound to be inconsistencies, misclassifications, missing data - in other words, errors, blunders, and possibly lies. These problems with individual items occur in any data set, but they are often hidden in a large mass of numbers even when these numbers are generated out of computer interactions." (David S Salsburg, "Errors, Blunders, and Lies: How to Tell the Difference", 2017)

"Missing data can pose more or less of a problem depending on how random they are. Missing values scattered haphazardly throughout your data set may be inconvenient, but they don’t necessarily pose a threat to your analyses. When missing data follow a clearer pattern, there may be a problem. In those cases, you might need to revisit how you collect your data." (Shonna D Watters et al, "The Practical Guide for HR Analytics: Using data to inform, transform, and empower HR decisions", 2019

"Unless we’re collecting data ourselves, there’s a limit to how much we can do to combat the problem of missing data. But we can and should remember to ask who or what might be missing from the data we’re being told about. Some missing numbers are obvious […]. Other omissions show up only when we take a close look at the claim in question." (Tim Harford, "The Data Detective: Ten easy rules to make sense of statistics", 2020)

"[Making reasoned macro calls] starts with having the best and longest-time-series data you can find. You may have to take some risks in terms of the quality of data sources, but it amazes me how people are often more willing to act based on little or no data than to use data that is a challenge to assemble." (Robert J Shiller)

🔭Data Science: Hallucinations (Just the Quotes)

"A hallucination is a fact, not an error; what is erroneous is a judgment based upon it." (Bertrand Russell, "On the Nature of Acquaintance: Neutral Monism", 1914)

"There is no theory we may hold and no observation we can make that will retain so much as its old defective reference to the facts if the net be altered. Tinitus, paraestheaias, hallucinations, delusions, confusions and disorientations intervene. Thus empiry confirms that if our nets are undefined, our facts are undefined, and to the 'real' we can attribute not so much as one quality or 'form'. With determi-nation of the net, the unknowable object of knowledge, the 'thing in itself', ceases to be unknowable." (Norbert Wiener, "Cybernetics: Or Control and Communication in the Animal and the Machine", 1948)

"Cyberspace. A consensual hallucination experienced daily by billions of legitimate operators, in every nation, by children being taught mathematical concepts. [...] A graphic representation of data abstracted from banks of every computer in the human system. Unthinkable complexity. Lines of light ranged in the nonspace of the mind, clusters and constellations of data." (William Gibson, "Neuromancer", 1984)

"I think we will get the hallucination problem to a much, much better place. I think it will take us a year and a half, two years. Something like that. But at that point we won't still talk about these. There's a balance between creativity and perfect accuracy, and the model will need to learn when you want one or the other." (Sam Altman, The Associated Press, 2023) [on Chatbot's hallucination] 

"Despite their impressive capabilities, LLMs are not without limitations. One of the most significant challenges is the problem of hallucination, where an LLM generates factually incorrect or misleading information that appears plausible. This is particularly problematic in domains requiring high factual accuracy, such as healthcare, finance, and legal applications. To mitigate hallucinations and enhance the reliability of LLM outputs,  Retrieval-Augmented Generation (RAG) has emerged as a powerful technique. RAG works by dynamically retrieving relevant information from an external knowledge source (such as a knowledge graph) at inference time, rather than just relying on pre-trained knowledge. This approach ensures that the model has access to up-to-date and accurate data, grounding answers in verified information rather than generating content purely from its internal representations." (Aldo Marzullo et al, "Graph Machine Learning" 2nd Ed., 2025)

"Generative AI tools for coding are sometimes inaccurate. They can produce results that look good but are wrong. This is common with LLMs. They can write code or chat like a person. And sometimes, they share information that’s just plain wrong. Not just a bit off, but totally backwards or nonsense. And they say it so confidently! We call this 'hallucinating', which is a funny term, but it makes sense." (Jeremy C Morgan, "Coding with AI: Examples in Python", 2025)

"LLMs are trained on large volumes of data, which inherently provides them with an immense knowledge base and understanding of different languages. Yet, LLMs at their core are complex text completion engines. Since this knowledge and understanding of language is compressed in a very high-dimensional latent space. LLMs end up using these in a very fluid and intelligible way (which often leads to hallucinations). In order to guide LLMs to focus on specific topics or pieces of information to solve certain tasks, (for instance, question-answering from a given piece of text), it is important to provide contextual information explicitly. While most current generations of LLMs have extremely wide context windows, it is recommended to preprocess context into overlapping smaller chunks for better results, reduced latency, and so on. For similar reasons, it is also recommended to preprocess contextual information in clear and task-specific formats. This aspect of context preprocessing is extremely useful in Retrieval-Gugmented Generation (RAG) scenarios." (Joseph Babcock & Raghav Bali, "Generative AI with Python and PyTorch" 2nd. Ed., 2025)

"Beyond intentionally misleading content, GenAI systems can produce inaccurate information unintentionally. LLMs are prone to hallucination, generating plausible but false statements with the same confidence as accurate ones. In enterprise contexts, this poses particular risks: an AI assistant might report incorrect financial figures, fabricate customer details, or misrepresent historical trends. Organizations deploying GenAI must implement validation mechanisms, human oversight, and retrieval-augmented approaches that ground model outputs in verified data sources." (Bennie Haelen, "ML and Generative AI in the Data Lakehouse Building and Deploying AI Applications at Scale", 2026)

"Ensuring that a language model reliably retrieves and presents correct information, often referred to as factual recall, is critical for any production-grade application. Whether you’re building an internal helpdesk assistant, a medical Q&A system, or an automated compliance auditor, users expect concise, accurate answers that align with up-to-date source material. Unfortunately, without explicit context, even the most powerful LLM can hallucinate or omit key facts." (Bennie Haelen, "ML and Generative AI in the Data Lakehouse Building and Deploying AI Applications at Scale", 2026)

"GenAI models are remarkably powerful, but they carry an inherent limitation rooted in how they are built. Every foundation model has a training cutoff, a point in time beyond which it has no awareness of world events, product changes, regulatory updates, or organizational developments. This knowledge gap is not a flaw that can be patched with better prompting. It’s a structural consequence of how models are trained, and it grows wider every day the model remains in production without retraining. Combined with the tendency to hallucinate when asked about topics outside their training distribution, models operating on stale knowledge can confidently deliver responses that are factually incorrect, dangerously outdated, or simply no longer relevant to the user’s context." (Bennie Haelen, "ML and Generative AI in the Data Lakehouse Building and Deploying AI Applications at Scale", 2026)

"[...] LLMs raise serious concerns about ethics, bias and fairness, errors in reasoning, hallucinations, and misuse (e.g., misinformation and disinformation). These concerns are exacerbated by modern LLMs being both literal and figurative 'black boxes': Literal black boxes because many advanced AI systems are proprietary and the weights (trained parameters of the models) are not released to the public; and figurative black boxes because even the open-source AI models are so complicated that understanding them and developing safety guardrails has thus far proven extremely difficult." (Mike X Cohen,"50 ML Projects To Understand LLMs", 2026)

"RAG applications must be built with semantics, metadata, and governance in mind. The retrieved information must be high-quality, secure, and appropriate for the user’s role. Equally important is monitoring and management: checking whether source data has changed, ensuring vector stores remain accurate, and watching for hallucinations or data leakage. Organizations are definitely starting to experiment with RAG models today; some are putting them into production applications. Some believe that using RAG helps mitigate hallucinations because it is grounded in trusted organizational data." (Fern Halper, "Data Makes the World Go 'Round", 2026)

More quotes on Hallucinations under Mind Literature

🔭Data Science: Signals (Just the Quotes)

"If statistical graphics, although born just yesterday, extends its reach every day, it is because it replaces long tables of numbers and it allows one not only to embrace at glance the series of phenomena, but also to signal the correspondences or anomalies, to find the causes, to identify the laws." (Émile Cheysson, cca. 1877)

"The term closed loop-learning process refers to the idea that one learns by determining what s desired and comparing what is actually taking place as measured at the process and feedback for comparison. The difference between what is desired and what is taking place provides an error indication which is used to develop a signal to the process being controlled." (Harold Chestnut, 1984) 

"Complexity is not an objective factor but a subjective one. Supersignals reduce complexity, collapsing a number of features into one. Consequently, complexity must be understood in terms of a specific individual and his or her supply of supersignals. We learn supersignals from experience, and our supply can differ greatly from another individual's. Therefore there can be no objective measure of complexity." (Dietrich Dorner, "The Logic of Failure: Recognizing and Avoiding Error in Complex Situations", 1989)

"An artificial neural network is an information-processing system that has certain performance characteristics in common with biological neural networks. Artificial neural networks have been developed as generalizations of mathematical models of human cognition or neural biology, based on the assumptions that: (1) Information processing occurs at many simple elements called neurons. (2) Signals are passed between neurons over connection links. (3) Each connection link has an associated weight, which, in a typical neural net, multiplies the signal transmitted. (4) Each neuron applies an activation function (usually nonlinear) to its net input (sum of weighted input signals) to determine its output signal." (Laurene Fausett, "Fundamentals of Neural Networks", 1994)

"Data are generally collected as a basis for action. However, unless potential signals are separated from probable noise, the actions taken may be totally inconsistent with the data. Thus, the proper use of data requires that you have simple and effective methods of analysis which will properly separate potential signals from probable noise." (Donald J Wheeler, "Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"No matter what the data, and no matter how the values are arranged and presented, you must always use some method of analysis to come up with an interpretation of the data.

While every data set contains noise, some data sets may contain signals. Therefore, before you can detect a signal within any given data set, you must first filter out the noise." (Donald J Wheeler," Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"While all data contain noise, some data contain signals. Before you can detect a signal, you must filter out the noise." (Donald J Wheeler, "Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"The most familiar example of swarm intelligence is the human brain. Memory, perception and thought all arise out of the nett actions of billions of individual neurons. As we saw earlier, artificial neural networks (ANNs) try to mimic this idea. Signals from the outside world enter via an input layer of neurons. These pass the signal through a series of hidden layers, until the result emerges from an output layer. Each neuron modifies the signal in some simple way. It might, for instance, convert the inputs by plugging them into a polynomial, or some other simple function. Also, the network can learn by modifying the strength of the connections between neurons in different layers." (David G Green, "The Serendipity Machine: A voyage of discovery through the unexpected world of computers", 2004)

"Data analysis is not generally thought of as being simple or easy, but it can be. The first step is to understand that the purpose of data analysis is to separate any signals that may be contained within the data from the noise in the data. Once you have filtered out the noise, anything left over will be your potential signals. The rest is just details." (Donald J Wheeler," Myths About Data Analysis", International Lean & Six Sigma Conference, 2012)

"Finding patterns is easy in any kind of data-rich environment […] The key is in determining whether the patterns represent signal or noise." (Nate Silver, "The Signal and the Noise: Why So Many Predictions Fail-but Some Don’t", 2012)

"The signal is the truth. The noise is what distracts us from the truth." (Nate Silver, "The Signal and the Noise: Why So Many Predictions Fail-but Some Don't", 2012)

"A signal is a useful message that resides in data. Data that isn’t useful is noise. […] When data is expressed visually, noise can exist not only as data that doesn’t inform but also as meaningless non-data elements of the display (e.g. irrelevant attributes, such as a third dimension of depth in bars, color variation that has no significance, and artificial light and shadow effects)." (Stephen Few, "Signal: Understanding What Matters in a World of Noise", 2015)

"To find signals in data, we must learn to reduce the noise - not just the noise that resides in the data, but also the noise that resides in us. It is nearly impossible for noisy minds to perceive anything but noise in data. […] Signals always point to something. In this sense, a signal is not a thing but a relationship. Data becomes useful knowledge of something that matters when it builds a bridge between a question and an answer. This connection is the signal." (Stephen Few, "Signal: Understanding What Matters in a World of Noise", 2015)

"Information theory leads to the quantification of the information content of the source, as denoted by entropy, the characterization of the information-bearing capacity of the communication channel, as related to its noise characteristics, and consequently the establishment of the relationship between the information content of the source and the capacity of the channel. In short, information theory provides a quantitative measure of the information contained in message signals and help determine the capacity of a communication system to transfer this information from source to sink over a noisy channel in a reliable fashion." (Ali Grami, "Information Theory", 2016)

 

More on Signals in Business Intelligence, Graphical Representation, Systems Engineering 

🔭Data Science: Regression toward the Mean (Just the Quotes)

"Whenever we make any decision based on the expectation that matters will return to 'normal', we are employing the notion of regression to the mean." (Peter L Bernstein, "Against the Gods: The Remarkable Story of Risk", 1996)

"Regression to the mean occurs when the process produces results that are statistically independent or negatively correlated. With strong negative serial correlation, extremes are likely to be reversed each time (which would reinforce the instructors' error). In contrast, with strong positive dependence, extreme results are quite likely to be clustered together." (Dan Trietsch, "Statistical Quality Control : A loss minimization approach", 1998) 

"Unfortunately, people are poor intuitive scientists, generally failing to reason in accordance with the principles of scientific method. For example, people do not generate sufficient alternative explanations or consider enough rival hypotheses. People generally do not adequately control for confounding variables when they explore a novel environment. People’s judgments are strongly affected by the frame in which the information is presented, even when the objective information is unchanged. People suffer from overconfidence in their judgments (underestimating uncertainty), wishful thinking (assessing desired outcomes as more likely than undesired outcomes), and the illusion of control (believing one can predict or influence the outcome of random events). People violate basic rules of probability, do not understand basic statistical concepts such as regression to the mean, and do not update beliefs according to Bayes’ rule. Memory is distorted by hindsight, the availability and salience of examples, and the desirability of outcomes. And so on."  (John D Sterman, "Business Dynamics: Systems thinking and modeling for a complex world", 2000)

 "People often attribute meaning to phenomena governed only by a regression to the mean, the mathematical tendency for an extreme value of an at least partially chance-dependent quantity to be followed by a value closer to the average. Sports and business are certainly chancy enterprises and thus subject to regression. So is genetics to an extent, and so very tall parents can be expected to have offspring who are tall, but probably not as tall as they are. A similar tendency holds for the children of very short parents." (John A Paulos, "A Mathematician Plays the Stock Market", 2003)

"'Regression to the mean' […] says that, in any series of events where chance is involved, very good or bad performances, high or low scores, extreme events, etc. tend on the average, to be followed by more average performance or less extreme events. If we do extremely well, we're likely to do worse the next time, while if we do poorly, we're likely to do better the next time. But regression to the mean is not a natural law. Merely a statistical tendency. And it may take a long time before it happens." (Peter Bevelin, "Seeking Wisdom: From Darwin to Munger",  2003)

"Another aspect of representativeness that is misunderstood or ignored is the tendency of regression to the mean. Stochastic phenomena where the outcomes vary randomly around stable values (so-called stationary processes) exhibit the general tendency that extreme outcomes are more likely to be followed by an outcome closer to the mean or mode than by other extreme values in the same direction. For example, even a bright student will observe that her or his performance in a test following an especially outstanding outcome tends to be less brilliant. Similarly, extremely low or extremely high sales in a given period tend to be followed by sales that are closer to the stable mean or the stable trend." (Hans G Daellenbach & Donald C McNickle, "Management Science: Decision making through systems thinking", 2005)

"Behavioural research shows that we tend to use simplifying heuristics when making judgements about uncertain events. These are prone to biases and systematic errors, such as stereotyping, disregard of sample size, disregard for regression to the mean, deriving estimates based on the ease of retrieving instances of the event, anchoring to the initial frame, the gambler’s fallacy, and wishful thinking, which are all affected by our inability to consider more than a few aspects or dimensions of any phenomenon or situation at the same time." (Hans G Daellenbach & Donald C McNickle, "Management Science: Decision making through systems thinking", 2005)

"Concluding that the population is becoming more centralized by observing behavior at the extremes is called the 'Regression to the Mean' Fallacy. […] When looking for a change in a population, do not look only at the extremes; there you will always find a motion to the mean. Look at the entire population." (Charles Livingston & Paul Voakes, "Working with Numbers and Statistics: A handbook for journalists", 2005)

"'Regression to the mean' describes a natural phenomenon whereby, after a short period of success, things tend to return to normal immediately afterwards. This notion applies particularly to random events." (Alan Graham, "Developing Thinking in Statistics", 2006)

"regression to the mean: The fact that unexpectedly high or low numbers from the mean are an exception and are usually followed by numbers that are closer to the mean. Over the long haul, we tend to get relatively more numbers that are near the mean compared to numbers that are far from the mean." (Hari Singh, "Framed! Solve an Intriguing Mystery and Master How to Make Smart Choices", 2006)

 "A naive interpretation of regression to the mean is that heights, or baseball records, or other variable phenomena necessarily become more and more 'average' over time. This view is mistaken because it ignores the error in the regression predicting y from x. For any data point xi, the point prediction for its yi will be regressed toward the mean, but the actual yi that is observed will not be exactly where it is predicted. Some points end up falling closer to the mean and some fall further." (Andrew Gelman & Jennifer Hill, "Data Analysis Using Regression and Multilevel/Hierarchical Models", 2007)

"Regression toward the mean. That is, in any series of random events an extraordinary event is most likely to be followed, due purely to chance, by a more ordinary one." (Leonard Mlodinow, "The Drunkard’s Walk: How Randomness Rules Our Lives", 2008)

"Regression does not describe changes in ability that happen as time passes […]. Regression is caused by performances fluctuating about ability, so that performances far from the mean reflect abilities that are closer to the mean." (Gary Smith, "Standard Deviations", 2014)

"We encounter regression in many contexts - pretty much whenever we see an imperfect measure of what we are trying to measure. Standardized tests are obviously an imperfect measure of ability. [...] Each experimental score is an imperfect measure of “ability,” the benefits from the layout. To the extent there is randomness in this experiment - and there surely is - the prospective benefits from the layout that has the highest score are probably closer to the mean than was the score." (Gary Smith, "Standard Deviations", 2014)

"When a trait, such as academic or athletic ability, is measured imperfectly, the observed differences in performance exaggerate the actual differences in ability. Those who perform the best are probably not as far above average as they seem. Nor are those who perform the worst as far below average as they seem. Their subsequent performances will consequently regress to the mean." (Gary Smith, "Standard Deviations", 2014)

"The term shrinkage is used in regression modeling to denote two ideas. The first meaning relates to the slope of a calibration plot, which is a plot of observed responses against predicted responses. When a dataset is used to fit the model parameters as well as to obtain the calibration plot, the usual estimation process will force the slope of observed versus predicted values to be one. When, however, parameter estimates are derived from one dataset and then applied to predict outcomes on an independent dataset, overfitting will cause the slope of the calibration plot (i.e., the shrinkage factor ) to be less than one, a result of regression to the mean. Typically, low predictions will be too low and high predictions too high. Predictions near the mean predicted value will usually be quite accurate. The second meaning of shrinkage is a statistical estimation method that preshrinks regression coefficients towards zero so that the calibration plot for new data will not need shrinkage as its calibration slope will be one." (Frank E. Harrell Jr., "Regression Modeling Strategies: With Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis" 2nd Ed, 2015)

"Often when people relate essentially the same variable in two different groups, or at two different times, they see this same phenomenon - the tendency of the response variable to be closer to the mean than the predicted value. Unfortunately, people try to interpret this by thinking that the performance of those far from the mean is deteriorating, but it’s just a mathematical fact about the correlation. So, today we try to be less judgmental about this phenomenon and we call it regression to the mean. We managed to get rid of the term 'mediocrity', but the name regression stuck as a name for the whole least squares fitting procedure - and that’s where we get the term regression line." (Richard D De Veaux et al, "Stats: Data and Models", 2016)

"Regression toward the mean is pervasive. In sports, excellent performance tends to be followed by good, but less outstanding, performance. [...] By contrast, the good news about regression toward the mean is that very poor performance tends to be followed by improved performance. If you got the worst score in your statistics class on the first exam, you probably did not do so poorly on the second exam (but you were probably still below the mean)." (Alan Agresti et al, Statistics: The Art and Science of Learning from Data" 4th Ed., 2018)

🔭Data Science: Asymmetry (Just the Quotes)

"Some distributions [...] are symmetrical about their central value. Other distributions have marked asymmetry and are said to be skew. Skew distributions are divided into two types. If the 'tail' of the distribution reaches out into the larger values of the variate, the distribution is said to show positive skewness; if the tail extends towards the smaller values of the variate, the distribution is called negatively skew." (Michael J Moroney,Facts from Figures", 1951)

“Rut seldom is asymmetry merely the absence of symmetry. Even in asymmetric designs one feels symmetry as the norm from which one deviates under the influence of forces of non-formal character.” (Hermann Weyl, “Symmetry”, 1952)

"If a distribution were perfectly symmetrical, all symmetry-plot points would be on the diagonal line. Off-line points indicate asymmetry. Points fall above the line when distance above the median is greater than corresponding distance below the median. A consistent run of above-the-line points indicates positive skew; a run of below-the-line points indicates negative skew." (Lawrence C Hamilton,Regression with Graphics: A second course in applied statistics", 1991)

“An asymmetry in the present is understood as having originated from a past symmetry.” (Michael Leyton, “Symmetry, Causality, Mind”, 1992)

"Chaos demonstrates that deterministic causes can have random effects […] There's a similar surprise regarding symmetry: symmetric causes can have asymmetric effects. […] This paradox, that symmetry can get lost between cause and effect, is called symmetry-breaking. […] From the smallest scales to the largest, many of nature's patterns are a result of broken symmetry; […]" (Ian Stewart & Martin Golubitsky,Fearful Symmetry: Is God a Geometer?", 1992)

“Approximate symmetry is a softening of the hard dichotomy between symmetry and asymmetry. The extent of deviation from exact symmetry that can still be considered approximate symmetry will depend on the context and the application and could very well be a matter of personal taste.” (Joe Rosen, “Symmetry Rules: How Science and Nature Are Founded on Symmetry”, 2008)

"[…] in cybernetics, control is seen not as a function of one agent over something else, but as residing within circular causal networks, maintaining stabilities in a system. Circularities have no beginning, no end and no asymmetries. The control metaphor of communication, by contrast, punctuates this circularity unevenly. It privileges the conceptions and actions of a designated controller by distinguishing between messages sent in order to cause desired effects and feedback that informs the controller of successes or failures." (Klaus Krippendorff,On Communicating: Otherness, Meaning, and Information", 2009)

“[…] asymmetry can be defined only relative to symmetry, and vice versa. Asymmetric elements in paintings or buildings are most effective when superimposed against a background of symmetry.” (Alan Lightman, “The Accidental Universe: The World You Thought You Knew”, 2014)

"The higher the dimension, in other words, the higher the number of possible interactions, and the more disproportionally difficult it is to understand the macro from the micro, the general from the simple units. This disproportionate increase of computational demands is called the curse of dimensionality." (Nassim N Taleb,Skin in the Game: Hidden Asymmetries in Daily Life", 2018)

"Many statistical procedures perform more effectively on data that are normally distributed, or at least are symmetric and not excessively kurtotic" (fat-tailed), and where the mean and variance are approximately constant. Observed time series frequently require some form of transformation before they exhibit these distributional properties, for in their 'raw' form they are often asymmetric." (Terence C Mills,Applied Time Series Analysis: A practical guide to modeling and forecasting", 2019)

🔭Data Science: Decision Trees (Just the Quotes)

"A good way to evaluate a model is to look at a visual representation of it. After all, what is easier to understand - a table full of mathematical relationships or a graphic displaying a decision tree with all of its splits and branches?" (Seth Paul et al. "Preparing and Mining Data with Microsoft SQL Server 2000 and Analysis", 2002)

"When generating trees, it is usually optimal to grow a larger tree than is justifiable and then prune it back. The main reason this works well is that stop splitting rules do not look far enough forward. That is, stop splitting rules tend to underfit, meaning that even if a rule stops at a split for which the next candidate splits give little improvement, it may be that splitting them one layer further will give a large improvement in accuracy." (Bertrand Clarke et al, "Principles and Theory for Data Mining and Machine Learning", 2009)

"A predictive model overfits the training set when at least some of the predictions it returns are based on spurious patterns present in the training data used to induce the model. Overfitting happens for a number of reasons, including sampling variance and noise in the training set. The problem of overfitting can affect any machine learning algorithm; however, the fact that decision tree induction algorithms work by recursively splitting the training data means that they have a natural tendency to segregate noisy instances and to create leaf nodes around these instances. Consequently, decision trees overfit by splitting the data on irrelevant features that only appear relevant due to noise or sampling variance in the training data. The likelihood of overfitting occurring increases as a tree gets deeper because the resulting predictions are based on smaller and smaller subsets as the dataset is partitioned after each feature test in the path." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)

"Decision trees are also discriminative models. Decision trees are induced by recursively partitioning the feature space into regions belonging to the different classes, and consequently they define a decision boundary by aggregating the neighboring regions belonging to the same class. Decision tree model ensembles based on bagging and boosting are also discriminative models." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)

"Decision trees are also considered nonparametric models. The reason for this is that when we train a decision tree from data, we do not assume a fixed set of parameters prior to training that define the tree. Instead, the tree branching and the depth of the tree are related to the complexity of the dataset it is trained on. If new instances were added to the dataset and we rebuilt the tree, it is likely that we would end up with a (potentially very) different tree." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)

"Decision trees are considered a good predictive model to start with, and have many advantages. Interpretability, variable selection, variable interaction, and the flexibility to choose the level of complexity for a decision tree all come into play." (Ralph Winters, "Practical Predictive Analytics", 2017)

"Decision trees are important for a few reasons. First, they can both classify and regress. It requires literally one line of code to switch between the two models just described, from a classification to a regression. Second, they are able to determine and share the feature importance of a given training set." (Russell Jurney, "Agile Data Science 2.0: Building Full-Stack Data Analytics Applications with Spark", 2017)

"Decision trees have two advantages over many of the algorithms [...]: the resulting model can easily be visualized and understood by nonexperts (at least for smaller trees), and the algorithms are completely invariant to scaling of the data. As each feature is processed separately, and the possible splits of the data don’t depend on scaling, no preprocessing like normalization or standardization of features is needed for decision tree algorithms. In particular, decision trees work well when you have features that are on completely different scales, or a mix of binary and continuous features. The main downside of decision trees is that even with the use of pre-pruning, they tend to overfit and provide poor generalization performance. Therefore, in most applications, the ensemble methods we discuss next are usually used in place of a single decision tree." (Andreas C Müller & Sarah Guido, "Introduction to Machine Learning with Python: A Guide for Data Scientists", 2017)

"Gradient boosted decision trees are among the most powerful and widely used models for supervised learning. Their main drawback is that they require careful tuning of the parameters and may take a long time to train. Similarly to other tree-based models, the algorithm works well without scaling and on a mixture of binary and continuous features. As with other tree-based models, it also often does not work well on high-dimensional sparse data." (Andreas C Müller & Sarah Guido, "Introduction to Machine Learning with Python: A Guide for Data Scientists", 2017)

 "The main reason why pruning tends to improve classification performance on future examples is that the removal of low-level tests, which have poor statistical support, usually reduces the danger of overfitting. This, however, works only up to a certain point. If overdone, a very high extent of pruning can (in the extreme) result in the decision being replaced with a single leaf labeled with the majority class." (Miroslav Kubat," An Introduction to Machine Learning" 2nd Ed., 2017)

"From a typical training set, many alternative decision trees can be created. As a rule, smaller trees are to be preferred, their main advantages being interpretability, removal of irrelevant and redundant attributes, and lower danger of overfitting noisy training data." (Miroslav Kubat, "An Introduction to Machine Learning" 2nd Ed., 2017)

"An advantage of random forests is that it works with both regression and classification trees so it can be used with targets whose role is binary, nominal, or interval. They are also less prone to overfitting than a single decision tree model. A disadvantage of a random forest is that they generally require more trees to improve their accuracy. This can result in increased run times, particularly when using very large data sets." (Richard V McCarthy et al, "Applying Predictive Analytics: Finding Value in Data", 2019)

"Decision trees show the breakdown of the data by one variable then another in a very intuitive way, though they are generally just diagrams that don’t actually encode data visually." (Robert Grant, "Data Visualization: Charts, Maps and Interactive Graphics", 2019)

"Random forests are essentially an ensemble of trees. They use many short trees, fitted to multiple samples of the data, and the predictions are averaged for each observation. This helps to get around a problem that trees, and many other machine learning techniques, are not guaranteed to find optimal models, in the way that linear regression is. They do a very challenging job of fitting non-linear predictions over many variables, even sometimes when there are more variables than there are observations. To do that, they have to employ 'greedy algorithms', which find a reasonably good model but not necessarily the very best model possible." (Robert Grant, "Data Visualization: Charts, Maps and Interactive Graphics", 2019)

"Decision tree is a graphical representation of a decision situation in which decision situation points (nodes) are connected together by arcs (one for each alternative on a decision) and terminate in ovals (the action that is the result of all the decisions made on the path leading to that oval). [...] A tree is made up of multilevel group of elements called nodes. A node is nothing more than a point at which subsidiary data originate. This particular logical data structure is called a tree simply because it looks like a tree, usually turned upside down. Genealogists use a schema called a tree to show ancestral descent of a person, family or group. Data associated by a tree schema are hierarchical. They branch from a point or node without forming loops or polygons. Data presented in a tree structure make two conditions. First, the tree must have a single root node. Second, all nodes other than the root node must be related to one and only one higher level node." (C S V Murthy, "Data and Businesss Analytics", 2020)

🔭Data Science: Evidence (Just the Quotes)

"Rule 1. Original data should be presented in a way that will preserve the evidence in the original data for all the predictions assumed to be useful." (Walter A Shewhart, "Economic Control of Quality of Manufactured Product", 1931)

"Rule 2. Any summary of a distribution of numbers in terms of symmetric functions should not give an objective degree of belief in any one of the inferences or predictions to be made therefrom that would cause human action significantly different from what this action would be if the original distributions had been taken as evidence." (Walter A Shewhart, "Economic Control of Quality of Manufactured Product", 1931)

"[...] there is evidence that significance tests have been a genuine block to achieving [...] knowledge." (Denton E Morrison & Ramon E Henkel, "Significance tests reconsidered", The American Sociologist 4, 1969)

"Confidence intervals give a feeling of the uncertainty of experimental evidence, and (very important) give it in the same units [...] as the original observations." (Mary G Natrella, "The relation between confidence intervals and tests of significance", American Statistician 14, 1960)

"The null-hypothesis significance test treats ‘acceptance’ or ‘rejection’ of a hypothesis as though these were decisions one makes. But a hypothesis is not something, like a piece of pie offered for dessert, which can be accepted or rejected by a voluntary physical action. Acceptance or rejection of a hypothesis is a cognitive process, a degree of believing or disbelieving which, if rational, is not a matter of choice but determined solely by how likely it is, given the evidence, that the hypothesis is true." (William W Rozeboom, "The fallacy of the null–hypothesis significance test", Psychological Bulletin 57, 1960)

"Scientific discovery, or the formulation of scientific theory, starts in with the unvarnished and unembroidered evidence of the senses. It starts with simple observation - simple, unbiased, unprejudiced, naive, or innocent observation - and out of this sensory evidence, embodied in the form of simple propositions or declarations of fact, generalizations will grow up and take shape, almost as if some process of crystallization or condensation were taking place. Out of a disorderly array of facts, an orderly theory, an orderly general statement, will somehow emerge." (Sir Peter B Medawar, "Is the Scientific Paper Fraudulent?", The Saturday Review, 1964)

"We have overwhelming evidence that available information plus analysis does not lead to knowledge. The management science team can properly analyse a situation and present recommendations to the manager, but no change occurs. The situation is so familiar to those of us who try to practice management science that I hardly need to describe the cases." (C West Churchman, "Managerial acceptance of scientific recommendations", California Management Review Vol 7, 1964)

"Science consists simply of the formulation and testing of hypotheses based on observational evidence; experiments are important where applicable, but their function is merely to simplify observation by imposing controlled conditions." (Henry L Batten, "Evolution of the Earth", 1971)

"Statistics is a body of methods and theory applied to numerical evidence in making decisions in the face of uncertainty." (Lawrence Lapin, "Statistics for Modern Business Decisions", 1973)

"The language of association and prediction is probably most often used because the evidence seems insufficient to justify a direct causal statement. A better practice is to state the causal hypothesis and then to present the evidence along with an assessment with respect to the causal hypothesis - instead of letting the quality of the data determine the language of the explanation." (Edward R Tufte, "Data Analysis for Politics and Policy", 1974)

"All interpretations made by a scientist are hypotheses, and all hypotheses are tentative. They must forever be tested and they must be revised if found to be unsatisfactory. Hence, a change of mind in a scientist, and particularly in a great scientist, is not only not a sign of weakness but rather evidence for continuing attention to the respective problem and an ability to test the hypothesis again and again." (Ernst Mayr, "The Growth of Biological Thought: Diversity, Evolution and Inheritance", 1982)

"It has been widely felt, probably for thirty years and more, that significance tests are overemphasized and often misused and that more emphasis should be put on estimation and prediction. While such a shift of emphasis does seem to be occurring, for example in medical statistics, the continued very extensive use of significance tests is on the one hand alarming and on the other evidence that they are aimed, even if imperfectly, at some widely felt need." (David R Cox, "Some general aspects of the theory of statistics", International Statistical Review 54, 1986)

"Like a detective, a data analyst will experience many dead ends, retrace his steps, and explore many alternatives before settling on a single description of the evidence in front of him." (David Lubinsky & Daryl Pregibon , "Data analysis as search", Journal of Econometrics Vol. 38 (1–2), 1988)

"Subjective probability, also known as Bayesian statistics, pushes Bayes' theorem further by applying it to statements of the type described as 'unscientific' in the frequency definition. The probability of a theory (e.g. that it will rain tomorrow or that parity is not violated) is considered to be a subjective 'degree of belief - it can perhaps be measured by seeing what odds the person concerned will offer as a bet. Subsequent experimental evidence then modifies the initial degree of belief, making it stronger or weaker according to whether the results agree or disagree with the predictions of the theory in question." (Roger J Barlow, "Statistics: A guide to the use of statistical methods in the physical sciences", 1989)

"Probability theory is an ideal tool for formalizing uncertainty in situations where class frequencies are known or where evidence is based on outcomes of a sufficiently long series of independent random experiments. Possibility theory, on the other hand, is ideal for formalizing incomplete information expressed in terms of fuzzy propositions." (George Klir, "Fuzzy sets and fuzzy logic", 1995)

"[…] the simplest hypothesis proposed as an explanation of phenomena is more likely to be the true one than is any other available hypothesis, that its predictions are more likely to be true than those of any other available hypothesis, and that it is an ultimate a priori epistemic principle that simplicity is evidence for truth." (Richard Swinburne, "Simplicity as Evidence for Truth", 1997)

"When significance tests are used and a null hypothesis is not rejected, a major problem often arises - namely, the result may be interpreted, without a logical basis, as providing evidence for the null hypothesis." (David F Parkhurst, "Statistical Significance Tests: Equivalence and Reverse Tests Should Reduce Misinterpretation", BioScience Vol. 51 (12), 2001)

"Essentially, the null hypothesis is that there is not a significant difference between two results. It will be seen that differences may have to be quite large in some instances before they are statistically significant, especially in the case of small data sets of high variability. Statistics will not say whether or not an apparent difference is real, but will only give the probability that it could have been as large as it is by chance alone. Often, the answer will be that there is no reason to believe a difference exists other than due to a chance occurrence, based on the statistical evidence available. Remember that this is not saying that there is no difference but that the evidence presented is insufficient to support the belief that the difference is not more than a random effect." (Cheryl Cihon & John K Taylor, "Statistical Techniques for Data Analysis" 2nd. ed., 2005)

"One cautious approach is represented by Bernoulli’s more conservative outlook. If there are very strong reasons for believing that an observation has suffered an accident that made the value in the data-file thoroughly untrustworthy, then reject it; in the absence of clear evidence that an observation, identified by formal rule as an outlier, is unacceptable then retain it unless there is lack of trust that the laboratory obtaining it is conscientiously operated by able persons who have [...] taken every care.'" (David Finney, "Calibration Guidelines Challenge Outlier Practices", The American Statistician Vol 60 (4), 2006)

"Scholars feel the need to present tables of model parameters in academic articles (perhaps just as evidence that they ran the analysis they claimed to have run), but these tables are rarely interpreted other than for their sign and statistical significance. Most of the numbers in these tables are never even discussed in the text. From the perspective of the applied data analyst, R packages without procedures to compute quantities of scientific interest are woefully incomplete. A better approach focuses on quantities of direct scientific interest rather than uninterpretable model parameters. [...] For each quantity of interest, the user needs some summary that includes a point estimate and a measure of uncertainty such as a standard error, confidence interval, or a distribution. The methods of calculating these differ greatly across theories of inference and methods of analysis. However, from the user’s perspective, the result is almost always the same: the point estimate and uncertainty of some quantity of interest." (Kousuke Imai et al, "Toward a Common Framework for Statistical Analysis and Development", Journal of Computational and Graphical Statistics vol. 17, 2008)

"Data analysis is careful thinking about evidence." (Michael Milton, "Head First Data Analysis", 2009)

"Data clusters are everywhere, even in random data. Someone who looks for an explanation will inevitably find one, but a theory that fits a data cluster is not persuasive evidence. The found explanation needs to make sense and it needs to be tested with uncontaminated data." (Gary Smith, "Standard Deviations", 2014)

"In general, when building statistical models, we must not forget that the aim is to understand something about the real world. Or predict, choose an action, make a decision, summarize evidence, and so on, but always about the real world, not an abstract mathematical world: our models are not the reality - a point well made by George Box in his oft-cited remark that "all models are wrong, but some are useful". (David Hand, "Wonderful examples, but let's not close our eyes", Statistical Science 29, 2014)

"The dialectical interplay of experiment and theory is a key driving force of modern science. Experimental data do only have meaning in the light of a particular model or at least a theoretical background. Reversely theoretical considerations may be logically consistent as well as intellectually elegant: Without experimental evidence they are a mere exercise of thought no matter how difficult they are. Data analysis is a connector between experiment and theory: Its techniques advise possibilities of model extraction as well as model testing with experimental data." (Achim Zielesny, "From Curve Fitting to Machine Learning" 2nd Ed., 2016)

"In terms of characteristics, a data scientist has an inquisitive mind and is prepared to explore and ask questions, examine assumptions and analyse processes, test hypotheses and try out solutions and, based on evidence, communicate informed conclusions, recommendations and caveats to stakeholders and decision makers." (Jesús Rogel-Salazar, "Data Science and Analytics with Python", 2017)

"With the growing availability of massive data sets and user-friendly analysis software, it might be thought that there is less need for training in statistical methods. This would be naïve in the extreme. Far from freeing us from the need for statistical skills, bigger data and the rise in the number and complexity of scientific studies makes it even more difficult to draw appropriate conclusions. More data means that we need to be even more aware of what the evidence is actually worth." (David Spiegelhalter, "The Art of Statistics: Learning from Data", 2019)

"The general principles of starting with a well-defined question, engaging in careful observation, and then formulating hypotheses and assessing the strength of evidence for and against them became known as the scientific method." (Michael Friendly & Howard Wainer, "A History of Data Visualization and Graphic Communication", 2021)

"Absence of evidence is not evidence of absence." (Martin Rees)

"The deepest sin of the human mind is to believe things without evidence." (Thomas H Huxley)

🔭Data Science: Time Series (Just the Quotes)

"No observations are absolutely trustworthy. In no field of observation can we entirely rule out the possibility that an observation is vitiated by a large measurement or execution error. If a reading is found to lie a very long way from its fellows in a series of replicate observations, there must be a suspicion that the deviation is caused by a blunder or gross error of some kind. [...] One sufficiently erroneous reading can wreck the whole of a statistical analysis, however many observations there are." (Francis J Anscombe, "Rejection of Outliers", Technometrics Vol. 2 (2), 1960)

"It is almost impossible to define 'time-sequence chart' in a clear and unambiguous manner because of the many forms and adaptations open to this type of chart. However. it might be said that, in essence, time-sequence chart portrays a chain of activities through time, indicates the type of activity in each link of the chain, shows clearly the position of the link in the total sequence chain, and indicates the duration of each activity. The time sequence chart may also contain verbal elements explaining when to begin an activity, how long to continue the activity, and a description of the activity. The chart may also indicate when to blend a given activity with another and the point at which a given activity is completed. The basic time-sequence chart may also be accompanied by verbal explanations and by secondary or contributory charts." (Cecil H Meyers, "Handbook of Basic Graphs: A modern approach", 1970)

"A time series is a sequence of observations, usually ordered in time, although in some cases the ordering may be according to another dimension. The feature of time series analysis which distinguishes it from other statistical analysis is the explicit recognition of the importance of the order in which the observations are made. While in many problems the observations are statistically independent, in time series successive observations may be dependent, and the dependence may depend on the positions in the sequence. The nature of a series and the structure of its generating process also may involve in other ways the sequence in which the observations are taken." (Theodore W Anderson, "The Statistical Analysis of Time Series", 1971)

"Entropy theory, on the other hand, is not concerned with the probability of succession in a series of items but with the overall distribution of kinds of items in a given arrangement." (Rudolf Arnheim, "Entropy and Art: An Essay on Disorder and Order", 1974)

"When the statistician looks at the outside world, he cannot, for example, rely on finding errors that are independently and identically distributed in approximately normal distributions. In particular, most economic and business data are collected serially and can be expected, therefore, to be heavily serially dependent. So is much of the data collected from the automatic instruments which are becoming so common in laboratories these days. Analysis of such data, using procedures such as standard regression analysis which assume independence, can lead to gross error. Furthermore, the possibility of contamination of the error distribution by outliers is always present and has recently received much attention. More generally, real data sets, especially if they are long, usually show inhomogeneity in the mean, the variance, or both, and it is not always possible to randomize." (George E P Box, "Some Problems of Statistics and Everyday Life", Journal of the American Statistical Association, Vol. 74 (365), 1979)

"An especially effective device for enhancing the explanatory power of time-series displays is to add spatial dimensions to the design of the graphic, so that the data are moving over space (in two or three dimensions) as well as over time. […] Occasionally graphics are belligerently multivariate, advertising the technique rather than the data." (Edward R Tufte, "The Visual Display of Quantitative Information", 1983)

"The bar graph and the column graph are popular because they are simple and easy to read. These are the most versatile of the graph forms. They can be used to display time series, to display the relationship between two items, to make a comparison among several items, and to make a comparison between parts and the whole (total). They do not appear to be as 'statistical', which is an advantage to those people who have negative attitudes toward statistics. The column graph shows values over time, and the bar graph shows values at a point in time. bar graph compares different items as of a specific time (not over time)." (Anker V Andersen, "Graphing Financial Information: How accountants can use graphs to communicate", 1983)

"The problem with time-series is that the simple passage of time is not a good explanatory variable: descriptive chronology is not causal explanation. There are occasional exceptions, especially when there is a clear mechanism that drives the Y-variable." (Edward R Tufte, "The Visual Display of Quantitative Information", 1983)

"The time-series plot is the most frequently used form of graphic design. With one dimension marching along to the regular rhythm of seconds, minutes, hours, days, weeks, months, years, centuries, or millennia, the natural ordering of the time scale gives this design a strength and efficiency of interpretation found in no other graphic arrangement." (Edward R Tufte, "The Visual Display of Quantitative Information", 1983)

"There are several uses for which the line graph is particularly relevant. One is for a series of data covering a long period of time. Another is for comparing several series on the same graph. A third is for emphasizing the movement of data rather than the amount of the data. It also can be used with two scales on the vertical axis, one on the right and another on the left, allowing different series to use different scales, and it can be used to present trends and forecasts." (Anker V Andersen, "Graphing Financial Information: How accountants can use graphs to communicate", 1983)

 "A connected graph is appropriate when the time series is smooth, so that perceiving individual values is not important. A vertical line graph is appropriate when it is important to see individual values, when we need to see short-term fluctuations, and when the time series has a large number of values; the use of vertical lines allows us to pack the series tightly along the horizontal axis. The vertical line graph, however, usually works best when the vertical lines emanate from a horizontal line through the center of the data and when there are no long-term trends in the data." (William S Cleveland, "The Elements of Graphing Data", 1985)

"A time series is a special case of the broader dependent-independent variable category. Time is the independent variable. One important property of most time series is that for each time point of the data there is only a single value of the dependent variable; there are no repeat measurements. Furthermore, most time series are measured at equally-spaced or nearly equally-spaced points in time." (William S Cleveland, "The Elements of Graphing Data", 1985)

"This transition from uncertainty to near certainty when we observe long series of events, or large systems, is an essential theme in the study of chance." (David Ruelle, "Chance and Chaos", 1991)

"System dynamics models are not derived statistically from time-series data. Instead, they are statements about system structure and the policies that guide decisions. Models contain the assumptions being made about a system. A model is only as good as the expertise which lies behind its formulation. A good computer model is distinguished from a poor one by the degree to which it captures the essence of a system that it represents. Many other kinds of mathematical models are limited because they will not accept the multiple-feedback-loop and nonlinear nature of real systems." (Jay W Forrester, "Counterintuitive Behavior of Social Systems", 1995)

"Like modeling, which involves making a static one-time prediction based on current information, time-series prediction involves looking at current information and predicting what is going to happen. However, with time-series predictions, we typically are looking at what has happened for some period back through time and predicting for some point in the future. The temporal or time element makes time-series prediction both more difficult and more rewarding. Someone who can predict the future based on what has occurred in the past can clearly have tremendous advantages over someone who cannot." (Joseph P Bigus,"Data Mining with Neural Networks: Solving business problems from application development to decision support", 1996)

"Many of the basic functions performed by neural networks are mirrored by human abilities. These include making distinctions between items (classification), dividing similar things into groups (clustering), associating two or more things (associative memory), learning to predict outcomes based on examples (modeling), being able to predict into the future (time-series forecasting), and finally juggling multiple goals and coming up with a good-enough solution (constraint satisfaction)." (Joseph P Bigus,"Data Mining with Neural Networks: Solving business problems from application development to decision support", 1996)

"Averages, ranges, and histograms all obscure the time-order for the data. If the time-order for the data shows some sort of definite pattern, then the obscuring of this pattern by the use of averages, ranges, or histograms can mislead the user. Since all data occur in time, virtually all data will have a time-order. In some cases this time-order is the essential context which must be preserved in the presentation." (Donald J Wheeler," Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"No comparison between two values can be global. A simple comparison between the current figure and some previous value and convey the behavior of any time series. […] While it is simple and easy to compare one number with another number, such comparisons are limited and weak. They are limited because of the amount of data used, and they are weak because both of the numbers are subject to the variation that is inevitably present in weak world data. Since both the current value and the earlier value are subject to this variation, it will always be difficult to determine just how much of the difference between the values is due to variation in the numbers, and how much, if any, of the difference is due to real changes in the process." (Donald J Wheeler, "Understanding Variation: The Key to Managing Chaos" 2nd Ed., 2000)

"Time-series forecasting is essentially a form of extrapolation in that it involves fitting a model to a set of data and then using that model outside the range of data to which it has been fitted. Extrapolation is rightly regarded with disfavour in other statistical areas, such as regression analysis. However, when forecasting the future of a time series, extrapolation is unavoidable." (Chris Chatfield, "Time-Series Forecasting" 2nd Ed, 2000)

"Comparing series visually can be misleading […]. Local variation is hidden when scaling the trends. We first need to make the series stationary (removing trend and/or seasonal components and/or differences in variability) and then compare changes over time. To do this, we log the series (to equalize variability) and difference each of them by subtracting last year’s value from this year’s value." (Leland Wilkinson, "The Grammar of Graphics" 2nd Ed., 2005)

"Prior to the discovery of the butterfly effect it was generally believed that small differences averaged out and were of no real significance. The butterfly effect showed that small things do matter. This has major implications for our notions of predictability, as over time these small differences can lead to quite unpredictable outcomes. For example, first of all, can we be sure that we are aware of all the small things that affect any given system or situation? Second, how do we know how these will affect the long-term outcome of the system or situation under study? The butterfly effect demonstrates the near impossibility of determining with any real degree of accuracy the long term outcomes of a series of events." (Elizabeth McMillan, Complexity, "Management and the Dynamics of Change: Challenges for practice", 2008)

"Regression toward the mean. That is, in any series of random events an extraordinary event is most likely to be followed, due purely to chance, by a more ordinary one." (Leonard Mlodinow, "The Drunkard’s Walk: How Randomness Rules Our Lives", 2008)

"A time-series plot (sometimes also called a time plot) is a simple graph of data collected over time that can be invaluable in identifying trends or patterns that might be of interest.A time-series plot can be constructed by thinking of the data set as a bivariate data set, where y is the variable observed and x is the time at which the observation was made. These (x, y) pairs are plotted as in a scatterplot. Consecutive observations are then connected by a line segment; this aids in spotting trends over time." (Roxy Peck et al, "Introduction to Statistics and Data Analysis" 4th Ed., 2012)

"Using random processes in our models allows economists to capture the variability of time series data, but it also poses challenges to model builders. As model builders, we must understand the uncertainty from two different perspectives. Consider first that of the econometrician, standing outside an economic model, who must assess its congruence with reality, inclusive of its random perturbations. An econometrician’s role is to choose among different parameters that together describe a family of possible models to best mimic measured real world time series and to test the implications of these models. I refer to this as outside uncertainty. Second, agents inside our model, be it consumers, entrepreneurs, or policy makers, must also confront uncertainty as they make decisions. I refer to this as inside uncertainty, as it pertains to the decision-makers within the model. What do these agents know? From what information can they learn? With how much confidence do they forecast the future? The modeler’s choice regarding insiders’ perspectives on an uncertain future can have significant consequences for each model’s equilibrium outcomes." (Lars P Hansen, "Uncertainty Outside and Inside Economic Models", [Nobel lecture] 2013)

"A key difference between a traditional statistical problems and a time series problem is that often, in time series, the errors are not independent." (DeWayne R Derryberry, "Basic data analysis for time series with R", 2014)

"Either a logarithmic or a square-root transformation of the data would produce a new series more amenable to fit a simple trigonometric model. It is often the case that periodic time series have rounded minima and sharp-peaked maxima. In these cases, the square root or logarithmic transformation seems to work well most of the time." (DeWayne R Derryberry, "Basic data analysis for time series with R", 2014)

 "The random element in most data analysis is assumed to be white noise - normal errors independent of each other. In a time series, the errors are often linked so that independence cannot be assumed (the last examples). Modeling the nature of this dependence is the key to time series." (DeWayne R Derryberry, "Basic data analysis for time series with R", 2014)

"With time series though, there is absolutely no substitute for plotting. The pertinent pattern might end up being a sharp spike followed by a gentle taper down. Or, maybe there are weird plateaus. There could be noisy spikes that have to be filtered out. A good way to look at it is this: means and standard deviations are based on the naïve assumption that data follows pretty bell curves, but there is no corresponding 'default' assumption for time series data (at least, not one that works well with any frequency), so you always have to look at the data to get a sense of what’s normal. [...] Along the lines of figuring out what patterns to expect, when you are exploring time series data, it is immensely useful to be able to zoom in and out." (Field Cady, "The Data Science Handbook", 2017)

"[Making reasoned macro calls] starts with having the best and longest-time-series data you can find. You may have to take some risks in terms of the quality of data sources, but it amazes me how people are often more willing to act based on little or no data than to use data that is a challenge to assemble." (Robert J Shiller)