How to Detect Publication Bias in Psychological Research

2019 ◽  
Vol 227 (4) ◽  
pp. 261-279 ◽  
Author(s):  
Frank Renkewitz ◽  
Melanie Keiner
Keyword(s):  
Publication Bias ◽  
Effect Size ◽  
Statistical Power ◽  
Type I Error ◽  
Psychological Research ◽  
Type I ◽  
True Effect Size ◽  
Questionable Research Practices ◽  
True Effect ◽  
Meta Analyses

Abstract. Publication biases and questionable research practices are assumed to be two of the main causes of low replication rates. Both of these problems lead to severely inflated effect size estimates in meta-analyses. Methodologists have proposed a number of statistical tools to detect such bias in meta-analytic results. We present an evaluation of the performance of six of these tools. To assess the Type I error rate and the statistical power of these methods, we simulated a large variety of literatures that differed with regard to true effect size, heterogeneity, number of available primary studies, and sample sizes of these primary studies; furthermore, simulated studies were subjected to different degrees of publication bias. Our results show that across all simulated conditions, no method consistently outperformed the others. Additionally, all methods performed poorly when true effect sizes were heterogeneous or primary studies had a small chance of being published, irrespective of their results. This suggests that in many actual meta-analyses in psychology, bias will remain undiscovered no matter which detection method is used.

scholarly journals How to detect publication bias in psychological research? A comparative evaluation of six statistical methods

2018 ◽  
Author(s):  
Frank Renkewitz ◽  
Melanie Keiner
Keyword(s):  
Publication Bias ◽  
Effect Size ◽  
Statistical Power ◽  
Detection Methods ◽  
Type I ◽  
Optimal Method ◽  
True Effect Size ◽  
True Effect ◽  
Size Heterogeneity ◽  
Meta Analyses

Publication biases and questionable research practices are assumed to be two of the main causes of low replication rates observed in the social sciences. Both of these problems do not only increase the proportion of false positives in the literature but can also lead to severely inflated effect size estimates in meta-analyses. Methodologists have proposed a number of statistical tools to detect and correct such bias in meta-analytic results. We present an evaluation of the performance of six of these tools in detecting bias. To assess the Type I error rate and the statistical power of these tools we simulated a large variety of literatures that differed with regard to underlying true effect size, heterogeneity, number of available primary studies and variation of sample sizes in these primary studies. Furthermore, simulated primary studies were subjected to different degrees of publication bias. Our results show that the power of the detection methods follows a complex pattern. Across all simulated conditions, no method consistently outperformed all others. Hence, choosing an optimal method would require knowledge about parameters (e.g., true effect size, heterogeneity) that meta-analysts cannot have. Additionally, all methods performed badly when true effect sizes were heterogeneous or primary studies had a small chance of being published irrespective of their results. This suggests, that in many actual meta-analyses in psychology bias will remain undiscovered no matter which detection method is used.

scholarly journals The Effect of Publication Bias on the Assessment of Heterogeneity

2017 ◽  
Author(s):  
Hilde Augusteijn ◽  
Robbie Cornelis Maria van Aert ◽  
Marcel A. L. M. van Assen
Keyword(s):  
Publication Bias ◽  
Effect Size ◽  
Type I Error ◽  
Meta Analysis ◽  
Error Rates ◽  
Population Heterogeneity ◽  
Type I ◽  
Monte Carlo Simulation Study ◽  
True Effect Size ◽  
True Effect

One of the main goals of meta-analysis is to test and estimate the heterogeneity of effect size. We examined the effect of publication bias on the Q-test and assessments of heterogeneity, as a function of true heterogeneity, publication bias, true effect size, number of studies, and variation of sample sizes. The expected values of heterogeneity measures H2 and I2 were analytically derived, and the power and the type I error rate of the Q-test were examined in a Monte-Carlo simulation study. Our results show that the effect of publication bias on the Q-test and assessment of heterogeneity is large, complex, and non-linear. Publication bias can both dramatically decrease and increase heterogeneity. Extreme homogeneity can occur even when the population heterogeneity is large. Particularly if the number of studies is large and population effect size is small, publication bias can cause both extreme type I error rates and power of the Q-test close to 0 or 1. We therefore conclude that the Q-test of homogeneity and heterogeneity measures H2 and I2 are generally not valid in assessing and testing heterogeneity when publication bias is present, especially when the true effect size is small and the number of studies is large. We introduce a web application, Q-sense, which can be used to assess the sensitivity of the Q-test to publication bias, and we apply it to two published meta-analysis. Meta-analytic methods should be enhanced in order to be able to deal with publication bias in their assessment and tests of heterogeneity.

scholarly journals Dissertation R.C.M. van Aert

2018 ◽  
Author(s):  
Robbie Cornelis Maria van Aert
Keyword(s):  
Publication Bias ◽  
Effect Size ◽  
Meta Analysis ◽  
Original Study ◽  
The Other ◽  
Effect Sizes ◽  
True Effect Size ◽  
True Effect ◽  
Meta Analyses ◽  
Q Statistic

More and more scientific research gets published nowadays, asking for statistical methods that enable researchers to get an overview of the literature in a particular research field. For that purpose, meta-analysis methods were developed that can be used for statistically combining the effect sizes from independent primary studies on the same topic. My dissertation focuses on two issues that are crucial when conducting a meta-analysis: publication bias and heterogeneity in primary studies’ true effect sizes. Accurate estimation of both the meta-analytic effect size as well as the between-study variance in true effect size is crucial since the results of meta-analyses are often used for policy making. Publication bias distorts the results of a meta-analysis since it refers to situations where publication of a primary study depends on its results. We developed new meta-analysis methods, p-uniform and p-uniform*, which estimate effect sizes corrected for publication bias and also test for publication bias. Although the methods perform well in many conditions, these and the other existing methods are shown not to perform well when researchers use questionable research practices. Additionally, when publication bias is absent or limited, traditional methods that do not correct for publication bias outperform p¬-uniform and p-uniform*. Surprisingly, we found no strong evidence for the presence of publication bias in our pre-registered study on the presence of publication bias in a large-scale data set consisting of 83 meta-analyses and 499 systematic reviews published in the fields of psychology and medicine. We also developed two methods for meta-analyzing a statistically significant published original study and a replication of that study, which reflects a situation often encountered by researchers. One method is a frequentist whereas the other method is a Bayesian statistical method. Both methods are shown to perform better than traditional meta-analytic methods that do not take the statistical significance of the original study into account. Analytical studies of both methods also show that sometimes the original study is better discarded for optimal estimation of the true effect size. Finally, we developed a program for determining the required sample size in a replication analogous to power analysis in null hypothesis testing. Computing the required sample size with the method revealed that large sample sizes (approximately 650 participants) are required to be able to distinguish a zero from a small true effect.Finally, in the last two chapters we derived a new multi-step estimator for the between-study variance in primary studies’ true effect sizes, and examined the statistical properties of two methods (Q-profile and generalized Q-statistic method) to compute the confidence interval of the between-study variance in true effect size. We proved that the multi-step estimator converges to the Paule-Mandel estimator which is nowadays one of the recommended methods to estimate the between-study variance in true effect sizes. Two Monte-Carlo simulation studies showed that the coverage probabilities of Q-profile and generalized Q-statistic method can be substantially below the nominal coverage rate if the assumptions underlying the random-effects meta-analysis model were violated.

scholarly journals Cluster Wild Bootstrapping to Handle Dependent Effect Sizes in Meta-Analysis with a Small Number of Studies

2021 ◽  
Author(s):  
Megha Joshi ◽  
James E Pustejovsky ◽  
S. Natasha Beretvas
Keyword(s):  
Effect Size ◽  
Type I Error ◽  
Meta Analysis ◽  
Error Rates ◽  
Small Sample ◽  
Type I ◽  
Hypothesis Tests ◽  
Type I Error Rates ◽  
Meta Analyses ◽  
Small Sample Correction

The most common and well-known meta-regression models work under the assumption that there is only one effect size estimate per study and that the estimates are independent. However, meta-analytic reviews of social science research often include multiple effect size estimates per primary study, leading to dependence in the estimates. Some meta-analyses also include multiple studies conducted by the same lab or investigator, creating another potential source of dependence. An increasingly popular method to handle dependence is robust variance estimation (RVE), but this method can result in inflated Type I error rates when the number of studies is small. Small-sample correction methods for RVE have been shown to control Type I error rates adequately but may be overly conservative, especially for tests of multiple-contrast hypotheses. We evaluated an alternative method for handling dependence, cluster wild bootstrapping, which has been examined in the econometrics literature but not in the context of meta-analysis. Results from two simulation studies indicate that cluster wild bootstrapping maintains adequate Type I error rates and provides more power than extant small sample correction methods, particularly for multiple-contrast hypothesis tests. We recommend using cluster wild bootstrapping to conduct hypothesis tests for meta-analyses with a small number of studies. We have also created an R package that implements such tests.

scholarly journals A Multi-faceted Mess: A Review of Statistical Power Analysis in Psychology Journal Articles

2019 ◽  
Author(s):  
Rob Cribbie ◽  
Nataly Beribisky ◽  
Udi Alter
Keyword(s):  
Sample Size ◽  
Effect Size ◽  
Power Analysis ◽  
Statistical Power ◽  
Type I Error ◽  
A Priori ◽  
Type I ◽  
Specific Level ◽  
Maximum Sample Size ◽  
Power Analyses

Many bodies recommend that a sample planning procedure, such as traditional NHST a priori power analysis, is conducted during the planning stages of a study. Power analysis allows the researcher to estimate how many participants are required in order to detect a minimally meaningful effect size at a specific level of power and Type I error rate. However, there are several drawbacks to the procedure that render it “a mess.” Specifically, the identification of the minimally meaningful effect size is often difficult but unavoidable for conducting the procedure properly, the procedure is not precision oriented, and does not guide the researcher to collect as many participants as feasibly possible. In this study, we explore how these three theoretical issues are reflected in applied psychological research in order to better understand whether these issues are concerns in practice. To investigate how power analysis is currently used, this study reviewed the reporting of 443 power analyses in high impact psychology journals in 2016 and 2017. It was found that researchers rarely use the minimally meaningful effect size as a rationale for the chosen effect in a power analysis. Further, precision-based approaches and collecting the maximum sample size feasible are almost never used in tandem with power analyses. In light of these findings, we offer that researchers should focus on tools beyond traditional power analysis when sample planning, such as collecting the maximum sample size feasible.

scholarly journals Has the evidence for moral licensing been inflated by publication bias?

2019 ◽  
Vol 3 ◽  
Author(s):  
Niclas Kuper ◽  
Antonia Bott
Keyword(s):  
Publication Bias ◽  
Effect Size ◽  
Empirical Work ◽  
True Effect Size ◽  
Replication Studies ◽  
Immoral Behavior ◽  
Moral Licensing ◽  
Self Image ◽  
Meta Analyses ◽  
Licensing Effects

Moral licensing describes the phenomenon that displaying moral behavior can lead to subsequent immoral behavior. This is usually explained by the idea that an initial moral act affirms the moral self-image and hence licenses subsequent immoral acts. Previous meta-analyses on moral licensing indicate significant overall effects of d> .30. However, several large replication studies have either not found the effect or reported a substantially smaller effect size. The present article investigated whether this can be attributed to publication bias. Datasets from two previous meta-analyses on moral licensing were compared and when necessary modified. The larger dataset was used for the present analyses. Using PET-PEESE and a three-parameter-selection-model (3-PSM), we found some evidence for publication bias. The adjusted effect sizes were reduced to d= -0.05, p= .64 and d= 0.18, p= .002, respectively. While the first estimate could be an underestimation, we also found indications that the second estimate might exaggerate the true effect size. It is concluded that both the evidence for and the size of moral licensing effects has likely been inflated by publication bias. Furthermore, our findings indicate that culture moderates the moral licensing effect. Recommendations for future meta-analytic and empirical work are given. Subsequent studies on moral licensing should be adequately powered and ideally pre-registered.  

scholarly journals Low statistical power in biomedical science: a review of three human research domains

2017 ◽  
Vol 4 (2) ◽  
pp. 160254 ◽  
Author(s):  
Estelle Dumas-Mallet ◽  
Katherine S. Button ◽  
Thomas Boraud ◽  
Francois Gonon ◽  
Marcus R. Munafò
Keyword(s):  
Effect Size ◽  
Statistical Power ◽  
Meta Analysis ◽  
Statistical Significance ◽  
Average Power ◽  
Biomedical Science ◽  
Significant Finding ◽  
Biomedical Sciences ◽  
True Effect Size ◽  
Meta Analyses

Studies with low statistical power increase the likelihood that a statistically significant finding represents a false positive result. We conducted a review of meta-analyses of studies investigating the association of biological, environmental or cognitive parameters with neurological, psychiatric and somatic diseases, excluding treatment studies, in order to estimate the average statistical power across these domains. Taking the effect size indicated by a meta-analysis as the best estimate of the likely true effect size, and assuming a threshold for declaring statistical significance of 5%, we found that approximately 50% of studies have statistical power in the 0–10% or 11–20% range, well below the minimum of 80% that is often considered conventional. Studies with low statistical power appear to be common in the biomedical sciences, at least in the specific subject areas captured by our search strategy. However, we also observe evidence that this depends in part on research methodology, with candidate gene studies showing very low average power and studies using cognitive/behavioural measures showing high average power. This warrants further investigation.

scholarly journals Cannons and sparrows II: the enhanced Bernoulli exact method for determining statistical significance and effect size in the meta-analysis of k 2 × 2 tables

2021 ◽  
Vol 18 (1) ◽  
Author(s):  
Lawrence M. Paul
Keyword(s):  
Effect Size ◽  
Type I Error ◽  
Meta Analysis ◽  
Statistical Significance ◽  
Statistical Error ◽  
Exact Method ◽  
Type I ◽  
Exact Test ◽  
Inverse Variance ◽  
Meta Analyses

Abstract Background The use of meta-analysis to aggregate the results of multiple studies has increased dramatically over the last 40 years. For homogeneous meta-analysis, the Mantel–Haenszel technique has typically been utilized. In such meta-analyses, the effect size across the contributing studies of the meta-analysis differs only by statistical error. If homogeneity cannot be assumed or established, the most popular technique developed to date is the inverse-variance DerSimonian and Laird (DL) technique (DerSimonian and Laird, in Control Clin Trials 7(3):177–88, 1986). However, both of these techniques are based on large sample, asymptotic assumptions. At best, they are approximations especially when the number of cases observed in any cell of the corresponding contingency tables is small. Results This research develops an exact, non-parametric test for evaluating statistical significance and a related method for estimating effect size in the meta-analysis of k 2 × 2 tables for any level of heterogeneity as an alternative to the asymptotic techniques. Monte Carlo simulations show that even for large values of heterogeneity, the Enhanced Bernoulli Technique (EBT) is far superior at maintaining the pre-specified level of Type I Error than the DL technique. A fully tested implementation in the R statistical language is freely available from the author. In addition, a second related exact test for estimating the Effect Size was developed and is also freely available. Conclusions This research has developed two exact tests for the meta-analysis of dichotomous, categorical data. The EBT technique was strongly superior to the DL technique in maintaining a pre-specified level of Type I Error even at extremely high levels of heterogeneity. As shown, the DL technique demonstrated many large violations of this level. Given the various biases towards finding statistical significance prevalent in epidemiology today, a strong focus on maintaining a pre-specified level of Type I Error would seem critical. In addition, a related exact method for estimating the Effect Size was developed.

scholarly journals Bayesian Hypothesis Testing and Estimation under the Marginalized Random-Effects Meta-Analysis Model

2020 ◽  
Author(s):  
Robbie Cornelis Maria van Aert ◽  
Joris Mulder
Keyword(s):  
Hypothesis Testing ◽  
Random Effects ◽  
Effect Size ◽  
Meta Analysis ◽  
Bayes Factors ◽  
Estimation Methods ◽  
True Effect Size ◽  
Bayesian Hypothesis Testing ◽  
True Effect ◽  
Meta Analyses

Meta-analysis methods are used to synthesize results of multiple studies on the same topic. The most frequently used statistical model in meta-analysis is the random-effects model containing parameters for the average effect, between-study variance in primary study's true effect size, and random effects for the study specific effects. We propose Bayesian hypothesis testing and estimation methods using the marginalized random-effects meta-analysis (MAREMA) model where the study specific true effects are regarded as nuisance parameters which are integrated out of the model. A flat prior distribution is placed on the overall effect size in case of estimation and a proper unit information prior for the overall effect size is proposed in case of hypothesis testing. For the between-study variance in true effect size, a proper uniform prior is placed on the proportion of total variance that can be attributed to between-study variability. Bayes factors are used for hypothesis testing that allow testing point and one-sided hypotheses. The proposed methodology has several attractive properties. First, the proposed MAREMA model encompasses models with a zero, negative, and positive between-study variance, which enables testing a zero between-study variance as it is not a boundary problem. Second, the methodology is suitable for default Bayesian meta-analyses as it requires no prior information about the unknown parameters. Third, the methodology can even be used in the extreme case when only two studies are available, because Bayes factors are not based on large sample theory. We illustrate the developed methods by applying it to two meta-analyses and introduce easy-to-use software in the R package BFpack to compute the proposed Bayes factors.

Sign in / Sign up

Export Citation Format

Share Document