Animal studies demonstrate particular difficulty in conducting bias-free research, investigators rarely synthesize existing research using modern methods, and have the added problem of translating results in one species to another (humans). Recent guidelines are described that attempt to improve the quality of animal research.

This chapter will help you decide whether a research project is better suited to either a single-equation or multiple-equation modeling framework. We explain these exogeneity assumptions and describe statistical tests to evaluate whether any particular assumption is consistent with the data. We begin by delineating the shortcomings of the standard exogeneity concepts used in cross-sectional analysis. We then introduce key terms that help unpack the various concepts and take a deep dive into weak, strong, and super exogeneity. We offer definitions and explain what can be learned from different analyses and assumptions. We also consider the relationships between these exogeneity assumptions and causal inference. We then discuss tests that help discern whether an exogeneity assumption is reasonable with stationary data and offer a strategy for assessing the plausibility of each exogeneity assumption based on theory, previous evidence, and empirical tests. Last, we illustrate the strategy with an example.

Bias in scientists themselves is discussed – is it in our DNA? I argue that it is; it had survival value early in our hominin history but is now a negative force. These cognitive biases must be acknowledged, understood in the context of our professional lives, and mitigated. Bias also operates within research teams, as groupthink, and this too must be managed. Confirmation bias is a dominant force but are we “lumpers or splitters.” How are data grouped or categorized, and how does “framing” influence our response to data, a total of 35 cognitive biases are described. Scientific fraud is rare and an entirely different problem but the impulse for it may share some of the same psychological roots.

The “play of chance” is described and sources of bias in how data are prepared for analysis (grouping and categorization) is a common source of bias. Multiple comparison bias is another frequently observed bias but problems with a companion bias – subgroup analysis – are less well known. Statistical analysis depends on having independent observations, or some method of accounting for dependency, and this too is a commonplace error and source of bias; it especially threatens meta-analyses. Bias is introduced when the wrong statistical unit is analyzed because “clustering” in the data has been ignored, reducing independency.

This chapter begins by describing the consequences of uncertainty about the univariate characteristics of the data for multiple regression analysis. Next, we describe the bounds approach to inference, a general inferential framework for dealing with uncertainty when the exact limiting distribution associated with a test statistic is unknown or difficult to determine. We then introduce a bounds approach to inference that is appropriate when there is uncertainty about the properties of both the outcome and (at least some) explanatory variables. This approach uses critical value bounds to evaluate the null hypothesis of no long-run relationship based on the long-run multiplier (LRM) for each independent variable in either the ADL, GECM, or restricted versions of these models. We also describe the ARDL bounds approach to inference developed by Pesaran, Shin, and Smith, which assumes that the outcome is known to contain a unit root but allows uncertainty about the properties of the explanatory variables. We compare the relative advantages of these frameworks and provide a practical guide to critical bounds approaches to inference with two illustrative examples.

All of the data needed to examine and model an epidemic are difficult to obtain with any accuracy, during a pandemic. This includes: case fatality, calculating the number of infections, estimating the effective reproduction number (R, how many additional cases will be infected by a single case), the incubation period (time from infection to symptoms), and the serial interval (time from start of symptoms in the infector to symptoms starting in the infectee). The COVID-19 pandemic is used to demonstrate these difficulties. Secondary health effects are an important consequence of pandemics and bias in these studies is discussed, as is pandemic modeling.

Despite being considered the most compelling single study design for attributing causation to observed associations, randomized controlled trials (RCTs) carry their own susceptibility to bias. Secure randomization procedures are necessary and the conduct of the RCT must be exemplary. How study drop-outs are managed, and who enters data analysis, can substantially influence the RCT result. Other aspects of patient care, such as co-interventions, must be carefully managed. Is outcome data complete for all patients, and do the trialists fully report all the RCTs hypothesized outcomes? Is “intent-to-treat” the primary analytic strategy?

In this concluding chapter, we offer some final thoughts on best practices for the conduct of time series analysis. The tools discussed in this text should be applied with as much precision as the data allow and the results from these analyses should be presented with maximum transparency. An honest presentation of your data, models, and decisions is crucial to effectively convey the results and contribute to knowledge production. After a brief synopsis of the book, we highlight various points of uncertainty in the modeling process and present some guidelines for transparency. The final section provides a brief overview of time series analysis topics not covered in this book. For each of these additional topics, we provide key references to serve as starting points for further exploration.

Systematic reviews and meta-analysis, particularly of randomized trials, are considered the highest quality of evidence supporting causal associations. But they are not immune to bias, bias in the included studies themselves and in the process of synthesizing studies and pooling data. This chapter considers methods for systematically reviewing a complete body of literature, deciding if the data are amenable to meta-analysis, and appropriately conducting such an analysis.

This chapter discusses biases that are of particular importance in the field of pharmacology, the most important of which is confounding by indication. How can researchers delineate those side effects owed to a drug from effects of the disease the drug is treating? A related bias occurs when early symptoms of disease are being treated by a drug that is later falsely implicated as causing the disease (protopathic bias). The adverse event reporting system (AERS) is often used to detect drug effects and one bias, the Weber effect, is reviewed.

Are people already at increased risk for disease more likely to be exposed to the risk factor of interest? Does closer observation of people with a disease lead to a false association? In retrospective studies, do people with a disease recall prior exposures more (or less) that healthier people? Are research interviewers a source of biased data collection? Confounding is an existential threat in biomedical research; here a second factor, which is associated with both the disease and the risk factor being studied, is an actual cause of the disease. If studies cannot fully control for the effect of the second risk factor, residual confounding will bias the risk estimate. Who participates and doesn’t participate in research is another source of bias. How diseases and risk factors are classified and categorized may introduce bias, and changing defined categories is yet another source of bias.