The Interpretation of Scholars' Interpretations of Confidence Intervals: Criticism, Replication, and Extension of Hoekstra et al. (2014)

P > .05: The incorrect interpretation of "not significant" results is a significant problem

Statistically nonsignificant (p > .05) results from a null hypothesis significance test (NHST) are often mistakenly interpreted as evidence that the null hypothesis is true-that there is "no effect" or "no difference." However, many of these results occur because the study had low statistical power to detect an effect. Power below 50% is common, in which case a result of no statistical significance is more likely to be incorrect than correct. The inference of "no effect" is not valid even if power is high. NHST assumes that the null hypothesis is true; p is the probability of the data under the assumption that there is no effect. A statistical test cannot confirm what it assumes. These incorrect statistical inferences could be eliminated if decisions based on p values were replaced by a biological evaluation of effect sizes and their confidence intervals. For a single study, the observed effect size is the best estimate of the population effect size, regardless of the p value. Unlike p values, confidence intervals provide information about the precision of the observed effect. In the biomedical and pharmacology literature, methods have been developed to evaluate whether effects are "equivalent," rather than zero, as tested with NHST. These methods could be used by biological anthropologists to evaluate the presence or absence of meaningful biological effects. Most of what appears to be known about no difference or no effect between sexes, between populations, between treatments, and other circumstances in the biological anthropology literature is based on invalid statistical inference.

Bayesian Estimation with Informative Priors is Indistinguishable from Data Falsification

Criticism of null hypothesis significance testing, confidence intervals, and frequentist statistics in general has evolved into advocacy of Bayesian analyses with informative priors for strong inference. This paper shows that Bayesian analysis with informative priors is formally equivalent to data falsification because the information carried by the prior can be expressed as the addition of fabricated observations whose statistical characteristics are determined by the parameters of the prior. This property of informative priors makes clear that only the use of non-informative, uniform priors in all types of Bayesian analyses is compatible with standards of research integrity. At the same time, though, Bayesian estimation with uniform priors yields point and interval estimates that are identical or nearly identical to those obtained with frequentist methods. At a qualitative level, frequentist and Bayesian outcomes have different interpretations but they are interchangeable when uniform priors are used. Yet, Bayesian interpretations require either the assumption that population parameters are random variables (which they are not) or an explicit acknowledgment that the posterior distribution (which is thus identical to the likelihood function except for a scale factor) only expresses the researcher's beliefs and not any information about the parameter of concern.

Concise, Simple, and Not Wrong: In Search of a Short-Hand Interpretation of Statistical Significance

One challenge when communicating science to practitioners and the general public is accurately representing statistical results. In particular, describing the meaning of statistical significance to a non-scientific audience is especially difficult given the technical nature of a correct definition. Correct interpretations of statistical significance can be unintuitive, nuanced, and use unfamiliar technical language. As a result, when researchers are tasked with providing short and understandable interpretations of statistical significance it can be tempting to default to convenient but incorrect interpretations. In the current paper, we offer a concise, simple, and correct interpretation of statistical significance that is suitable for communications targeting a general audience.

Perceived Statistical Knowledge Level and Self-Reported Statistical Practice Among Academic Psychologists

Introduction: Publications arguing against the null hypothesis significance testing (NHST) procedure and in favor of good statistical practices have increased. The most frequently mentioned alternatives to NHST are effect size statistics (ES), confidence intervals (CIs), and meta-analyses. A recent survey conducted in Spain found that academic psychologists have poor knowledge about effect size statistics, confidence intervals, and graphic displays for meta-analyses, which might lead to a misinterpretation of the results. In addition, it also found that, although the use of ES is becoming generalized, the same thing is not true for CIs. Finally, academics with greater knowledge about ES statistics presented a profile closer to good statistical practice and research design. Our main purpose was to analyze the extension of these results to a different geographical area through a replication study.

Methods: For this purpose, we elaborated an on-line survey that included the same items as the original research, and we asked academic psychologists to indicate their level of knowledge about ES, their CIs, and meta-analyses, and how they use them. The sample consisted of 159 Italian academic psychologists (54.09% women, mean age of 47.65 years). The mean number of years in the position of professor was 12.90 (SD = 10.21).

Results: As in the original research, the results showed that, although the use of effect size estimates is becoming generalized, an under-reporting of CIs for ES persists. The most frequent ES statistics mentioned were Cohen's d and R²/η², which can have outliers or show non-normality or violate statistical assumptions. In addition, academics showed poor knowledge about meta-analytic displays (e.g., forest plot and funnel plot) and quality checklists for studies. Finally, academics with higher-level knowledge about ES statistics seem to have a profile closer to good statistical practices.

Conclusions: Changing statistical practice is not easy.This change requires statistical training programs for academics, both graduate and undergraduate.

Statistical Rituals: The Replication Delusion and How We Got There

The “replication crisis” has been attributed to misguided external incentives gamed by researchers (the strategic-game hypothesis). Here, I want to draw attention to a complementary internal factor, namely, researchers’ widespread faith in a statistical ritual and associated delusions (the statistical-ritual hypothesis). The “null ritual,” unknown in statistics proper, eliminates judgment precisely at points where statistical theories demand it. The crucial delusion is that the p value specifies the probability of a successful replication (i.e., 1 – p), which makes replication studies appear to be superfluous. A review of studies with 839 academic psychologists and 991 students shows that the replication delusion existed among 20% of the faculty teaching statistics in psychology, 39% of the professors and lecturers, and 66% of the students. Two further beliefs, the illusion of certainty (e.g., that statistical significance proves that an effect exists) and Bayesian wishful thinking (e.g., that the probability of the alternative hypothesis being true is 1 – p), also make successful replication appear to be certain or almost certain, respectively. In every study reviewed, the majority of researchers (56%–97%) exhibited one or more of these delusions. Psychology departments need to begin teaching statistical thinking, not rituals, and journal editors should no longer accept manuscripts that report results as “significant” or “not significant.”

A Cross-Sectional Analysis of Students' Intuitions When Interpreting CIs

We explored how students interpret the relative likelihood of capturing a population parameter at various points of a CI in two studies. First, an online survey of 101 students found that students' beliefs about the probability curve within a CI take a variety of shapes, and that in fixed choice tasks, 39% CI [30, 48] of students' responses deviated from true distributions. For open ended tasks, this proportion rose to 85%, 95% CI [76, 90]. We interpret this as evidence that, for many students, intuitions about CIs distributions are ill-formed, and their responses are highly susceptible to question format. Many students also falsely believed that there is substantial change in likelihood at the upper and lower limits of the CI, resembling a cliff effect (Rosenthal and Gaito, 1963; Nelson et al., 1986). In a follow-up study, a subset of 24 post-graduate students participated in a 45-min semi-structured interview discussing the students' responses to the survey. Analysis of interview transcripts identified several competing intuitions about CIs, and several new CI misconceptions. During the interview, we also introduced an interactive teaching program displaying a cat's eye CI, that is, a CI that uses normal distributions to depict the correct likelihood distribution. Cat's eye CIs were designed to help students understand likelihood distributions and the relationship between interval length, C% level and sample size. Observed changes in students' intuitions following this teaching program suggest that a brief intervention using cat's eyes can reduce CI misconceptions and increase accurate CI intuitions.