Category Archives: Uncategorized

Democracy and Citizens’ Happiness

For 30 years, I have been interested in cultural differences. I maintained a database of variables that vary across cultures, starting with Hofestede’s seminal rankings of 40 nations. Finding interesting variables was difficult and time consuming. The world has changed. Today it is easy to find interesting data on happiness, income, or type of government. Statistical software is also free (project R). This has changed the social sciences. Nowadays, the new problem is that data can be analyzed in many ways and that results can be inconclusive. As a result, social scientists can disagree even when the analyze the same data. Here I focus on predictors of national differences in happiness.

Happiness has been defined in many ways and any conclusion about national differences in happiness depends on the definition of happiness. The most widely used definition of happiness in the social sciences is subjective well-being. Accordingly, individuals define for themselves what they consider to be a good life and evaluate how close their actual lives are to their ideal lives. The advantage of this concept of well-being is that it does not impose values on the concept of happiness. Individuals in democratic countries could evaluate their lives based on different criteria than individuals in non-democratic countries. Thus, subjective well-being is not biased in favor of democracy, even though subjective conceptions of happiness emerged along with democracy in Western countries.

The most widely used measure of subjective well-being is Cantril’s ladder. Participants rate their lives on a scale from 0 = worst possible life to 10 = best possible life. This measure leaves it to participants to define what the worst or best possible life it. The best possible life in Denmark could be a very different life than the best possible life in Zimbabwe. Ratings on Cantril’s ladder are imperfect measures of subjective well-being and could distort comparisons of countries, but these ratings are currently used to compare the happiness of over 100 countries (WHR).

The Economist’s Intelligence Unit (EUI) has created ratings of countries’ forms of government that provides a measure of democracy (Democracy Index). Correlating the 2020 happiness means of countries with the democracy index produces a strong (linear) correlation of r = .68 (rank correlation r = .71).

This finding has been used to argue that democracies are better societies because they provide more happiness for their citizens (Williamson, 2022).

So the eastward expansion of democracy isn’t some US-led conspiracy to threaten Russia; it reflects the fact that, when given the choice, citizens tend to choose democracy and hope over autocracy and fear. They know instinctively that it brings a greater chance for happiness.

Although I am more than sympathetic to this argument, I am more doubtful that democracy alone is sufficient to produce more happiness. A strong correlation between democracy and happiness is insufficient to make this argument. It is well known that many predictors of nations’ happiness scores are strongly corelated with each other. One well known predictor is nations’ wealth or purchasing power. Money does buy essential goods. The best predictor of happiness is the median income per person that reflects the spending power of average citizens and is not distorted by international trade or rich elites.

While it is known that purchasing power is a predictor of well-being, it is often ignored how strong the relationship is. The linear correlation across nations is r = .79 (rank r = .82). It is often argued that the relationship between income is not linear and that money is more important in poorer countries. However, the correlation with log income is only slightly higher, r = .83.

This might suggest that purchasing power and democracy are both important for happiness. However, purchasing power and democracy are also strongly correlated, (linear r = .72, rank = .75). Multiple regression analysis can be used to see whether both variables independently contribute to the prediction of happiness.

Of course, dollars cannot be directly compared to ratings on a democracy index. To make the results comparable, I scored both variables from 0 for the lowest possible score to 1 for the highest possible score. For purchasing power, this variable ranged from Madagascar ($398) to Luxembourg ($26,321). For democracy, this variable ranged from Myanmar (1.02) to Norway (9.75).

The results show that purchasing power is a much stronger predictor of happiness than democracy.

The model predicts that a country with the lowest standing on purchasing power and democracy has a score of 3.63 on Cantril’s happiness measure. Increasing wealth to the maximum level without changing democracy would increase happiness to 3.63 + 3.13 = 6.76. In contrast, keeping purchasing power at the lowest level and increasing democracy to the highest level would increase happiness only to 3.63 + 0.48 = 4.11. One problem with statistical analyses across nations is that the sample size is limited by the number of nations. As a result, the positive relationship with democracy is not statistically significant and it is possible that the true effect is zero. In contrast, the effect of purchasing power is highly significant and it is unlikely (less than 5%) that the increase is less than 2.5 points.

Do these results imply that democracy is not very important for citizens’ happiness? Not necessarily. A regression analysis ignores the correlation between the predictor variables. It is possible that the correlation between purchasing power and democracy reflects at least in part a causal effect of democracy on wealth. For example, democratic governments may invest more in education and innovation and achieve higher economic growth. Democracies may also produce better working conditions and policies that benefit the working class rather than wealthy elites.

I will not repeat the mistake of many other social scientists to end with a strong conclusion that fits their world views based on weak and inconclusive data. The main aim of this blog post is to warn readers that social science is much more complicated than the natural sciences. Follow the science makes a lot of sense, when large clinical trials show strong effectiveness of drugs or vaccines. The social sciences can provide valuable information, but do not provide simple rules that can be followed to increase human well-being. This does not mean that social science is irrelevant. Ideally, social scientists would provide factual information and leave the interpretation to educated consumers.

Interpreting discrepancies between Self-Perceptions and IAT scores: Who is defensive?

In 1998, Anthony G. Greenwald and colleagues introduced the Implicit Association Test. Since then, Implicit Association Tests have been used in thousands of studies with millions of participants to study stereotypes and attitudes. The most prominent and controversial use of the race IAT that has been used to argue that many White Americans have more negative attitudes towards African Americans than they admit to others or even to themselves.

The popularity of IATs can be attributed to the use of IATs on the Project Implicit website that provides visitors of the website with the opportunity to take an IAT and to receive feedback about their performance. Over 1 million visitors have received feedback about their performance on the race IAT (Howell, Gaither, & Ratliff, 2015).

Providing participants with performance feedback can be valuable and educational. Coaches provide feedback to athletes so that they can improve their performance, and professors provide feedback about performance during midterms so that students can improve their performance on finals. However, the value of feedback depends on the accuracy of the feedback. As psychological researchers know, providing participants with false feedback is unethical and requires extensive debriefing to justify the use of false feedback in research. it is therefore crucial to examine the accuracy of performance feedback on the race IAT.

At face value, IAT feedback is objective and reflects participants’ responses to the stimuli that were presented during an IAT. However, this performance feedback should come with a warning that performance could vary across repeated administration of a test. For example, the retest reliability of performance on the race IAT has been estimated to be between r = .2 and r = .5. Even using a value of r = .5 implies that there is only a 75% probability that somebody with a score above average receives a score above average again on a second test (Rosenthal and Rubin, 1982).

However, the Project Implicit website gives the false impression that performance on IATs is rather consistent, while avoiding quantitative information about reliability.

FAQ5


Unreliability is not the only reason why performance feedback on the Project Implicit website could be misleading. Another problem is that visitors may be given the impression that performance on the race IAT reveals something about themselves that goes beyond performance on this specific task. One possible interpretation of race IAT scores is that they reveal implicit attitudes or evaluations of Black and White Americans. These implicit attitudes can be different from attitudes that individuals think they have that are called explicit attitudes. In fact, Greenwald et al. (1998) introduced IATs as a method that can detect implicit attitudes that can differ from explicit attitudes and this dual-attitude model has fueled interest in IATs.

The Project Implicit website does not provide a clear explanation of what Implicit association Tests test. Regarding the race IAT, visitors are told that it is not a measure of prejudice, but that it does measure their biases, even if these biases are not endorsed or contradict conscious beliefs.

FAQ11

However, other frequently asked question implies that IATs measure implicit stereotypes and attitudes. One question is how IATs measure implicit attitudes, implying that it can measure implicit attitudes (and that implicit attitudes exist).

FAQ2

Another one implies that performance on the race IAT reveals implicit attitudes that reflect cultural biases.

In short, while Project Implicit may not provide a clear explanation of what is being tested with an Implicit Association Test, it is strongly implied that test performance reveals something about participants’ racial biases that may contradict their self-perceptions.

An article by Howell, Gaither, and Ratliff (2015) makes this assumption explicit. This article examines how visitors of the Project Implicit website respond to performance feedback on the race IAT. The key claim of this article is that “people are generally defensive in response to feedback indicating that their implicit attitudes differ from their explicit attitudes” (p. 373). This statement rests on two assumptions. First, it makes the assumption of dual-attitude models that there are explicit and implicit attitudes, as suggested by Greenwald et al. (1998). Second, it implies that performance on a single race IAT provides highly valid information about implicit attitudes. These assumptions are necessary to place researchers in the position of an expert that know individuals’ implicit attitudes, just like a psychoanalyst is in a superior position to understand the true meaning of a dream. If test takers reject the truth, they are considered defensive because they are unwilling to accept the truth.

To measure defensiveness, Howell et al. (2015) used answers to three questions after visitors of the Project Implicit website received performance feedback on the race IAT, namely
(a) the IAT does not reflect anything about my thoughts or feelings unconscious or otherwise,
(b) whether I like my IAT score or not, it captures something important about me (reversed)
(c) the IAT reflects something about my automatic thoughts and feelings concerning this topic (reversed). Responses were made on a 1 = strongly disagree to 4 = strongly agree. On this scale, a score of 2.5 would imply neither agreement nor disagreement with the aforementioned statements.

There was hardly any difference in defensiveness scores between White (M = 2.31, SD = 0.68) Black (M = 2.38, SD = 0.74) or biracial (M = 2.33, SD = 0.73) participants. For White participants, a larger pro-White discrepancy was correlated with higher defensiveness scores, partial r = .16. The same result was found for Black participants, partial r = .13. A similar trend emerged for biracial participants. While these correlations are weak, they suggest that all three racial groups were less likely to believe in the accuracy of the feedback when the IAT scores showed a stronger pro-White bias than the self-ratings implied.

Howell et al. (2015) interpret these results as evidence of defensiveness. Accordingly, “White individuals want to avoid appearing racist (O’Brien et al., 2010) and Black individuals value pro-Black bias (Sniderman & Piazza, 2002)” (p. 378). However, this interpretation of the results rests on the assumption that the race IAT is an unbiased measure of racial attitudes. Howell et al. (2015) ignore a plausible alternative explanation of their results. The alternative explanation is that performance feedback on the race IAT is biased in favor of pro-White attitudes. One source of this bias could be the scoring of IATs which relies on the assumption that neutral attitudes correspond to a zero score. This assumption has been challenged in numerous articles (e.g., Blanton, Jaccard, Strauts, Mitchell, & Tetlock, 2015). It is also noteworthy that other implicit measures of racial attitudes show different results than the race IAT (Judd et al., 1995; Schimmack & Howard, 2021). Another problem is that there is little empirical support for dual-attitude models (Schimmack, 2021). Thus, it is impossible for IAT scores to provide truthful information that is discrepant from individuals’ self-knowledge (Schimmack, 2021).

Of course, people are defensive when they are confronted with unpleasant information and inconvenient truths. A prime example of defensiveness is the response of the researchers behind Project Implicit to valid scientific criticism of their interpretation of IAT scores.

About us

Despite several inquires about questionable or even misleading statements on the frequently asked question page, Project Implicit visitors are not informed that the wider scientific community has challenged the interpretation of performance feedback on the race IAT as valid information about individuals implicit attitudes. The simple fact that a single IAT score provides insufficient information to make valid claims about an individuals’ attitudes or behavioral tendencies is missing. Visitors should be informed that the most plausible and benign reason for a discrepancy between their test scores and their beliefs is that test scores could be biased. However, Project Implicit is unlikely to provide visitors with this information because the website is used for research purposes and willingness to participate in research might decrease when participants are told the truth about the mediocre validity of IATs.

Proponents of IATs often argue that taking an IAT can be educational. However, Howell et al. (2015) point out that even this alleged benefit is elusive because individuals are more likely to believe themselves than the race IAT feedback. Thus, rejection of IAT feedback, whether it is based on defensiveness or valid concerns about the validity of the test results, might undermine educational programs that aim to reduce actual racial biases. It is therefore problematic to use the race IAT in education and intervention programs.

2021 Replicability Report for the Psychology Department at the University of Amsterdam

Since 2011, it is an open secret that many published results in psychology journals do not replicate. The replicability of published results is particularly low in social psychology (Open Science Collaboration, 2015).

A key reason for low replicability is that researchers are rewarded for publishing as many articles as possible without concerns about the replicability of the published findings. This incentive structure is maintained by journal editors, review panels of granting agencies, and hiring and promotion committees at universities.

To change the incentive structure, I developed the Replicability Index, a blog that critically examined the replicability, credibility, and integrity of psychological science. In 2016, I created the first replicability rankings of psychology departments (Schimmack, 2016). Based on scientific criticisms of these methods, I have improved the selection process of articles to be used in departmental reviews.

1. I am using Web of Science to obtain lists of published articles from individual authors (Schimmack, 2022). This method minimizes the chance that articles that do not belong to an author are included in a replicability analysis. It also allows me to classify researchers into areas based on the frequency of publications in specialized journals. Currently, I cannot evaluate neuroscience research. So, the rankings are limited to cognitive, social, developmental, clinical, and applied psychologists.

2. I am using department’s websites to identify researchers that belong to the psychology department. This eliminates articles that are from other departments.

3. I am only using tenured, active professors. This eliminates emeritus professors from the evaluation of departments. I am not including assistant professors because the published results might negatively impact their chances to get tenure. Another reason is that they often do not have enough publications at their current university to produce meaningful results.

Like all empirical research, the present results rely on a number of assumptions and have some limitations. The main limitations are that
(a) only results that were found in an automatic search are included
(b) only results published in 120 journals are included (see list of journals)
(c) published significant results (p < .05) may not be a representative sample of all significant results
(d) point estimates are imprecise and can vary based on sampling error alone.

These limitations do not invalidate the results. Large difference in replicability estimates are likely to predict real differences in success rates of actual replication studies (Schimmack, 2022).

University of Amsterdam

The University of Amsterdam is the highest ranked European psychology department (QS Rankings). I used the department website to find core members of the psychology department. I found 48 senior faculty members. Not all researchers conduct quantitative research and report test statistics in their result sections. Therefore, the analysis is limited to 25 faculty members that had at least 100 test statistics.

A search of the database retrieved 13,529 test statistics. This is the highest number of statistical tests for all departments examined so far (Department Rankings). This partially explains the high ranking of the University of Amsterdam in rankings of prestige.

Figure 1 shows the z-curve plot for these results. I use the Figure to explain how a z-curve analysis provides information about replicability and other useful meta-statistics.

1. All test-statistics are converted into absolute z-scores as a common metric of the strength of evidence (effect size over sampling error) against the null-hypothesis (typically H0 = no effect). A z-curve plot is a histogram of absolute z-scores in the range from 0 to 6. The 2,034 z-scores greater than 6 are not shown because z-scores of this magnitude are extremely unlikely to occur when the null-hypothesis is true (particle physics uses z > 5 for significance). Although they are not shown, they are included in the computation of the meta-statistics.

2. Visual inspection of the histogram shows a drop in frequencies at z = 1.96 (solid red line) that corresponds to the standard criterion for statistical significance, p = .05 (two-tailed). This shows that published results are selected for significance. The dashed red line shows significance for p < .10, which is often used for marginal significance. Thus, there are more results that are presented as significant than the .05 criterion suggests.

3. To quantify the amount of selection bias, z-curve fits a statistical model to the distribution of statistically significant results (z > 1.96). The grey curve shows the predicted values for the observed significant results and the unobserved non-significant results. The statistically significant results (including z > 6) make up 35% of the total area under the grey curve. This is called the expected discovery rate because the results provide an estimate of the percentage of significant results that researchers actually obtain in their statistical analyses. In comparison, the percentage of significant results (including z > 6) includes 70% of the published results. This percentage is called the observed discovery rate, which is the rate of significant results in published journal articles. The difference between a 70% ODR and a 35% EDR provides an estimate of the extent of selection for significance. The difference of ~35 percentage points is large in absolute terns, but relatively small in comparison to other psychology departments. The upper level of the 95% confidence interval for the EDR is 46%. Thus, the discrepancy is not just random. To put this result in context, it is possible to compare it to the average for 120 psychology journals in 2010 (Schimmack, 2022). The ODR (70% vs. 72%) is similar, but the EDR is higher (35% vs. 28%), suggesting less severe selection for significance by faculty members at the University of Amsterdam that are included in this analysis.

4. The z-curve model also estimates the average power of the subset of studies with significant results (p < .05, two-tailed). This estimate is called the expected replication rate (ERR) because it predicts the percentage of significant results that are expected if the same analyses were repeated in exact replication studies with the same sample sizes. The ERR of 66% suggests a fairly high replication rate. The problem is that actual replication rates are lower than the ERR predictions (about 40% Open Science Collaboration, 2015). The main reason is that it is impossible to conduct exact replication studies and that selection for significance will lead to a regression to the mean when replication studies are not exact. Thus, the ERR represents the best case scenario that is unrealistic. In contrast, the EDR represents the worst case scenario in which selection for significance does not select more powerful studies and the success rate of replication studies is not different from the success rate of original studies. The EDR of 35% is below the actual replication success rate of 40%. To predict the success rate of actual replication studies, I am using the average of the EDR and ERR, which is called the actual replication prediction (ARP). For the University of Amsterdam, the ARP is (70 +35)/2 = 53%. This is somewhat higher than the currently best estimate of the success rate for actual replication studies based on the Open Science Collaboration project (~40%). Thus, research from the University of Amsterdam is expected to replicate at a higher rate than the replication rate for psychology in general.

5. The EDR can be used to estimate the risk that published results are false positives (i.e., a statistically significant result when H0 is true), using Soric’s (1989) formula for the maximum false discovery rate. An EDR of 35% implies that no more than 10% of the significant results are false positives, but the lower limit of the 95%CI of the EDR, 23%, allows for 18% false positive results. One solution to this problem is to lower the conventional criterion for statistical significance (Benjamin et al., 2017). Figure 2 shows that alpha = .005 reduces the point estimate of the FDR to 2% with an upper limit of the 95% confidence interval of 4%. Thus, without any further information readers could use this criterion to interpret results published in articles by psychology researchers at Western University.

Some researchers have changed research practices in response to the replication crisis. It is therefore interesting to examine whether replicability of newer research has improved. It is particularly interesting to examine changes at the University of Amsterdam because Erik-Jan Wagenmakers, a faculty member in the Methodology department, is a prominent advocate of methodological reforms. To examine this question, I performed a z-curve analysis for articles published in the past five year (2016-2021).

The results are disappointing. There is no evidence that research practices have changed in response to concerns about replication failures. The EDR estimate dropped from 35% to 25%, although this is not a statistically significant change. The ERR also decreased slightly from 72% to 69%. Therefore, the predicted success rate for actual replication studies decreased from 51% to 47%. This means that the University of Amsterdam decreased in rankings that focus on the past five years because some other departments have improved.

The replication crisis has been most severe in social psychology (Open Science Collaboration, 2015) and was in part triggered by concerns about social psychological research in the Netherlands. I therefore also conducted a z-curve analysis for the 10 faculty members in social psychology. The EDR is lower (24% vs. 35%) than for the whole department, which also implies a lower actual replication rate and a higher false positive risk.

There is variability across individual researchers, although confidence intervals are often wide due to the smaller number of test statistics. The table below shows the meta-statistics of all faculty members that provided results for the departmental z-curve. You can see the z-curve for individual faculty member by clicking on their name.

Rank  NameARPEDRERRFDR
1Jaap M. J. Murre7781742
2Hilde M. Geurts7376692
3Timo Stein7376702
4Hilde M. Huizenga6875613
5Maurits W. van der Molen6572574
6Astrid C. Homan6269554
7Wouter van den Bos6074476
8Frenk van Harreveld5464447
9Gerben A. van Kleef5370379
10K. Richard Ridderinkhof5369369
11Bruno Verschuere52723211
12Maartje E. J. Raijmakers51742813
13Merel Kindt48623510
14Mark Rotteveel47593410
15Sanne de Wit47742022
16Susan M. Bogels44632615
17Matthijs Baas44622516
18Arnoud R. Arntz43681726
19Filip van Opstal43652020
20Suzanne Oosterwijk42562913
21Edwin A. J. van Hooft40651530
22E. J. B. Doosje38611531
23Nils B. Jostmann37482615
24Barbara Nevicka37591433
25Reinout W. Wiers36472515

2021 Replicability Report for the Psychology Department at Western University

Since 2011, it is an open secret that many published results in psychology journals do not replicate. The replicability of published results is particularly low in social psychology (Open Science Collaboration, 2015).

A key reason for low replicability is that researchers are rewarded for publishing as many articles as possible without concerns about the replicability of the published findings. This incentive structure is maintained by journal editors, review panels of granting agencies, and hiring and promotion committees at universities.

To change the incentive structure, I developed the Replicability Index, a blog that critically examined the replicability, credibility, and integrity of psychological science. In 2016, I created the first replicability rankings of psychology departments (Schimmack, 2016). Based on scientific criticisms of these methods, I have improved the selection process of articles to be used in departmental reviews.

1. I am using Web of Science to obtain lists of published articles from individual authors (Schimmack, 2022). This method minimizes the chance that articles that do not belong to an author are included in a replicability analysis. It also allows me to classify researchers into areas based on the frequency of publications in specialized journals. Currently, I cannot evaluate neuroscience research. So, the rankings are limited to cognitive, social, developmental, clinical, and applied psychologists.

2. I am using department’s websites to identify researchers that belong to the psychology department. This eliminates articles that are from other departments.

3. I am only using tenured, active professors. This eliminates emeritus professors from the evaluation of departments. I am not including assistant professors because the published results might negatively impact their chances to get tenure. Another reason is that they often do not have enough publications at their current university to produce meaningful results.

Like all empirical research, the present results rely on a number of assumptions and have some limitations. The main limitations are that
(a) only results that were found in an automatic search are included
(b) only results published in 120 journals are included (see list of journals)
(c) published significant results (p < .05) may not be a representative sample of all significant results
(d) point estimates are imprecise and can vary based on sampling error alone.

These limitations do not invalidate the results. Large difference in replicability estimates are likely to predict real differences in success rates of actual replication studies (Schimmack, 2022).

Western University

I used the department website to find core members of the psychology department. I found 35 faculty members at the associate (9) or full professor (26) level. Not all researchers conduct quantitative research and report test statistics in their result sections. Therefore, the analysis is limited to 14 faculty members that had at least 100 test statistics.

Figure 1 shows the z-curve for all 6,080 tests statistics. I use the Figure to explain how a z-curve analysis provides information about replicability and other useful meta-statistics.

1. All test-statistics are converted into absolute z-scores as a common metric of the strength of evidence (effect size over sampling error) against the null-hypothesis (typically H0 = no effect). A z-curve plot is a histogram of absolute z-scores in the range from 0 to 6. The 865 z-scores greater than 6 are not shown because z-scores of this magnitude are extremely unlikely to occur when the null-hypothesis is true (particle physics uses z > 5 for significance). Although they are not shown, they are included in the computation of the meta-statistics.

2. Visual inspection of the histogram shows a drop in frequencies at z = 1.96 (solid red line) that corresponds to the standard criterion for statistical significance, p = .05 (two-tailed). This shows that published results are selected for significance. The dashed red line shows significance for p < .10, which is often used for marginal significance. Thus, there are more results that are presented as significant than the .05 criterion suggests.

3. To quantify the amount of selection bias, z-curve fits a statistical model to the distribution of statistically significant results (z > 1.96). The grey curve shows the predicted values for the observed significant results and the unobserved non-significant results. The statistically significant results (including z > 6) make up 38% of the total area under the grey curve. This is called the expected discovery rate because the results provide an estimate of the percentage of significant results that researchers actually obtain in their statistical analyses. In comparison, the percentage of significant results (including z > 6) includes 70% of the published results. This percentage is called the observed discovery rate, which is the rate of significant results in published journal articles. The difference between a 70% ODR and a 38% EDR provides an estimate of the extent of selection for significance. The difference of ~30 percentage points is large in absolute terns, but relatively small in comparison to other psychology departments. The upper level of the 95% confidence interval for the EDR is 56%. Thus, the discrepancy is not just random. To put this result in context, it is possible to compare it to the average for 120 psychology journals in 2010 (Schimmack, 2022). The ODR (70% vs. 72%) is similar, but the EDR is higher (38% vs. 28%), suggesting less severe selection for significance for research published by faculty members at Western University included in this analysis.

4. The z-curve model also estimates the average power of the subset of studies with significant results (p < .05, two-tailed). This estimate is called the expected replication rate (ERR) because it predicts the percentage of significant results that are expected if the same analyses were repeated in exact replication studies with the same sample sizes. The ERR of 70% suggests a fairly high replication rate. The problem is that actual replication rates are lower than the ERR predictions (about 40% Open Science Collaboration, 2015). The main reason is that it is impossible to conduct exact replication studies and that selection for significance will lead to a regression to the mean when replication studies are not exact. Thus, the ERR represents the best case scenario that is unrealistic. In contrast, the EDR represents the worst case scenario in which selection for significance does not select more powerful studies and the success rate of replication studies is not different from the success rate of original studies. The EDR of 38% is closer to the actual replication success rate of 40%. To predict the success rate of actual replication studies, I am using the average of the EDR and ERR, which is called the actual replication prediction (ARP). For Western University, the ARP is (70 +38)/2 = 54%. This is close to the currently best estimate of the success rate for actual replication studies based on the Open Science Collaboration project (~40%). Thus, research from Western University is expected to replicate at the average rate of actual replication studies.

5. The EDR can be used to estimate the risk that published results are false positives (i.e., a statistically significant result when H0 is true), using Soric’s (1989) formula for the maximum false discovery rate. An EDR of 38% implies that no more than 9% of the significant results are false positives, but the lower limit of the 95%CI of the EDR, 23%, allows for 18% false positive results. One solution to this problem is to lower the conventional criterion for statistical significance (Benjamin et al., 2017). Figure 2 shows that alpha = .005 reduces the point estimate of the FDR to 1% with an upper limit of the 95% confidence interval of 3%. Thus, without any further information readers could use this criterion to interpret results published in articles by psychology researchers at Western University.

Some researchers have changed research practices in response to the replication crisis. It is therefore interesting to examine whether replicability of newer research has improved. To examine this question, I performed a z-curve analysis for articles published in the past five year (2016-2021).

The results are amazing. The EDR increased to 70% with a relatively tight confidence interval ranging from 61% to 78%. The confidence interval does not overlap with the confidence interval for all-time z-scores. This makes Western only the second department that shows a statistically significant improvement in response to the replication crisis. Moreover, The ARP of 74% is the highest and much higher than some other departments.

There is variability across individual researchers, although confidence intervals are often wide due to the smaller number of test statistics. The table below shows the meta-statistics of all faculty members that provided results for the departmental z-curve. You can see the z-curve for individual faculty member by clicking on their name.

Rank  NameARPEDRERRFDR
1Rachel M. Calogero8088712
2Melvyn A. Goodale7478702
3Daniel Ansari6777584
4John Paul Minda6475545
5Stefan Kohler6366603
6Ryan A. Stevenson5765505
7Debra J. Jared54763311
8Erin A. Heerey5264418
9Stephen J. Lupker51693311
10Ken McRae51693211
11Ingrid S. Johnsrude49752218
12Lorne Campbell45622913
13Marc F. Joanisse45672417
14Victoria M. Esses39522615

2021 Replicability Report for the Psychology Department at McGill University

Since 2011, it is an open secret that many published results in psychology journals do not replicate. The replicability of published results is particularly low in social psychology (Open Science Collaboration, 2015).

A key reason for low replicability is that researchers are rewarded for publishing as many articles as possible without concerns about the replicability of the published findings. This incentive structure is maintained by journal editors, review panels of granting agencies, and hiring and promotion committees at universities.

To change the incentive structure, I developed the Replicability Index, a blog that critically examined the replicability, credibility, and integrity of psychological science. In 2016, I created the first replicability rankings of psychology departments (Schimmack, 2016). Based on scientific criticisms of these methods, I have improved the selection process of articles to be used in departmental reviews.

1. I am using Web of Science to obtain lists of published articles from individual authors (Schimmack, 2022). This method minimizes the chance that articles that do not belong to an author are included in a replicability analysis. It also allows me to classify researchers into areas based on the frequency of publications in specialized journals. Currently, I cannot evaluate neuroscience research. So, the rankings are limited to cognitive, social, developmental, clinical, and applied psychologists.

2. I am using department’s websites to identify researchers that belong to the psychology department. This eliminates articles that are from other departments.

3. I am only using tenured, active professors. This eliminates emeritus professors from the evaluation of departments. I am not including assistant professors because the published results might negatively impact their chances to get tenure. Another reason is that they often do not have enough publications at their current university to produce meaningful results.

Like all empirical research, the present results rely on a number of assumptions and have some limitations. The main limitations are that
(a) only results that were found in an automatic search are included
(b) only results published in 120 journals are included (see list of journals)
(c) published significant results (p < .05) may not be a representative sample of all significant results
(d) point estimates are imprecise and can vary based on sampling error alone.

These limitations do not invalidate the results. Large difference in replicability estimates are likely to predict real differences in success rates of actual replication studies (Schimmack, 2022).

McGill University

I used the department website to find core members of the psychology department. I found (only) 20 faculty members at the associate (5) or full professor (15) level. The reason is that McGill is going through a phase of renewal and currently has a large number of assistant professors before tenure that are not included in these analyses (14). It will be interesting to see the replicability of research at McGill in five years when these assistant professors are promoted to the rank of associate professor.

Not all researchers conduct quantitative research and report test statistics in their result sections. Therefore, the analysis is limited to 10 faculty members that had at least 100 significant test statistics. Thus, the results are by no means representative of the whole department with 34 faculty members, but I had to follow the same criteria that I used for other departments.

Figure 1 shows the z-curve for all 3,000 tests statistics. This is a relatively small number of z-scores. Larger departments and departments with more prolific researchers can have over 10,000 test statistics. I use the Figure to explain how a z-curve analysis provides information about replicability and other useful meta-statistics.

1. All test-statistics are converted into absolute z-scores as a common metric of the strength of evidence (effect size over sampling error) against the null-hypothesis (typically H0 = no effect). A z-curve plot is a histogram of absolute z-scores in the range from 0 to 6. The 412 z-scores greater than 6 are not shown because z-scores of this magnitude are extremely unlikely to occur when the null-hypothesis is true (particle physics uses z > 5 for significance). Although they are not shown, they are included in the computation of the meta-statistics.

2. Visual inspection of the histogram shows a steep drop in frequencies at z = 1.96 (solid red line) that corresponds to the standard criterion for statistical significance, p = .05 (two-tailed). This shows that published results are selected for significance. The dashed red line shows significance for p < .10, which is often used for marginal significance. Thus, there are more results that are presented as significant than the .05 criterion suggests.

3. To quantify the amount of selection bias, z-curve fits a statistical model to the distribution of statistically significant results (z > 1.96). The grey curve shows the predicted values for the observed significant results and the unobserved non-significant results. The statistically significant results (including z > 6) make up 21% of the total area under the grey curve. This is called the expected discovery rate because the results provide an estimate of the percentage of significant results that researchers actually obtain in their statistical analyses. In comparison, the percentage of significant results (including z > 6) includes 78% of the published results. This percentage is called the observed discovery rate, which is the rate of significant results in published journal articles. The difference between a 78% ODR and a 21% EDR provides an estimate of the extent of selection for significance. The difference of nearly 60 percentage points is the largest difference observed for any department analyzed so far (k = 11). The upper level of the 95% confidence interval for the EDR is 34%. Thus, the discrepancy is not just random. To put this result in context, it is possible to compare it to the average for 120 psychology journals in 2010 (Schimmack, 2022). The ODR (78% vs. 72%) is higher and the EDR (21% vs. 28%) is lower, suggesting more severe selection for significance for research published by McGill faculty members included in this analysis.

4. The z-curve model also estimates the average power of the subset of studies with significant results (p < .05, two-tailed). This estimate is called the expected replication rate (ERR) because it predicts the percentage of significant results that are expected if the same analyses were repeated in exact replication studies with the same sample sizes. The ERR of 66% suggests a fairly high replication rate. The problem is that actual replication rates are lower than the ERR predictions (about 40% Open Science Collaboration, 2015). The main reason is that it is impossible to conduct exact replication studies and that selection for significance will lead to a regression to the mean when replication studies are not exact. Thus, the ERR represents the best case scenario that is unrealistic. In contrast, the EDR represents the worst case scenario in which selection for significance does not select more powerful studies and the success rate of replication studies is not different from the success rate of original studies. The EDR of 21% is lower than the actual replication success rate of 40%. To predict the success rate of actual replication studies, I am using the average of the EDR and ERR, which is called the actual replication prediction (ARP). For Columbia, the ARP is (66 + 21)/2 = 44%. This is close to the currently best estimate of the success rate for actual replication studies based on the Open Science Collaboration project (~40%). Thus, research from McGill University is expected to replicate at the average rate of actual replication studies.

5. The EDR can be used to estimate the risk that published results are false positives (i.e., a statistically significant result when H0 is true), using Soric’s (1989) formula for the maximum false discovery rate. An EDR of 21% implies that no more than 20% of the significant results are false positives, but the lower limit of the 95%CI of the EDR, 12%, allows for 39% false positive results. Most readers are likely to agree that this is too high. One solution to this problem is to lower the conventional criterion for statistical significance (Benjamin et al., 2017). Figure 2 shows that alpha = .005 reduces the point estimate of the FDR to 3% with an upper limit of the 95% confidence interval of 9%. Thus, without any further information readers could use this criterion to interpret results published in articles by psychology researchers at Columbia. Of course, this criterion will be inappropriate for some researchers, but the present results show that the traditional alpha criterion of .05 is also inappropriate to maintain a reasonably low probability of false positive results.

Some researchers have changed research practices in response to the replication crisis. It is therefore interesting to examine whether replicability of newer research has improved. To examine this question, I performed a z-curve analysis for articles published in the past five year (2016-2021).

The results are disappointing. The point estimate, 18%, is even lower than for all year, 21%, although the difference could just be sampling error. Mostly, these results suggest that the psychology department at McGill University has not responded to the replication crisis in psychology, despite a low replication rate that provides more room for improvement. It will be interesting to see whether the large cohort of assistant professors adopted better research practices and will boost McGill’s standing in the replicability rankings of psychology departments.

There is considerable variability across individual researchers, although confidence intervals are often wide due to the smaller number of test statistics. The table below shows the meta-statistics of all faculty members that provided results for the departmental z-curve. You can see the z-curve for individual faculty member by clicking on their name.

Rank  NameARPEDRERRFDR
1Caroline Palmer7175682
2Kristine H. Onishi6069515
3Melanie A. Dirks54783112
4Richard Koestner50673211
5Yitzchak M. Binik49732516
6John E. Lydon44612814
7Jelena Ristic44691824
8Mark W. Baldwin43543311
9Jennifer A. Bartz39581922
10Blaine Ditto1727767

2021 Replicability Report for the Psychology Department at Princeton University

Since 2011, it is an open secret that many published results in psychology journals do not replicate. The replicability of published results is particularly low in social psychology (Open Science Collaboration, 2015).

A key reason for low replicability is that researchers are rewarded for publishing as many articles as possible without concerns about the replicability of the published findings. This incentive structure is maintained by journal editors, review panels of granting agencies, and hiring and promotion committees at universities.

To change the incentive structure, I developed the Replicability Index, a blog that critically examined the replicability, credibility, and integrity of psychological science. In 2016, I created the first replicability rankings of psychology departments (Schimmack, 2016). Based on scientific criticisms of these methods, I have improved the selection process of articles to be used in departmental reviews.

1. I am using Web of Science to obtain lists of published articles from individual authors (Schimmack, 2022). This method minimizes the chance that articles that do not belong to an author are included in a replicability analysis. It also allows me to classify researchers into areas based on the frequency of publications in specialized journals. Currently, I cannot evaluate neuroscience research. So, the rankings are limited to cognitive, social, developmental, clinical, and applied psychologists.

2. I am using department’s websites to identify researchers that belong to the psychology department. This eliminates articles that are from other departments.

3. I am only using tenured, active professors. This eliminates emeritus professors from the evaluation of departments. I am not including assistant professors because the published results might negatively impact their chances to get tenure. Another reason is that they often do not have enough publications at their current university to produce meaningful results.

Like all empirical research, the present results rely on a number of assumptions and have some limitations. The main limitations are that
(a) only results that were found in an automatic search are included
(b) only results published in 120 journals are included (see list of journals)
(c) published significant results (p < .05) may not be a representative sample of all significant results
(d) point estimates are imprecise and can vary based on sampling error alone.

These limitations do not invalidate the results. Large difference in replicability estimates are likely to predict real differences in success rates of actual replication studies (Schimmack, 2022).

Princeton University

I used the department website to find core members of the psychology department. I found 24 professors and 4 associate professors. I used Web of Science to download references related to the authors name and initial. An r-script searched for related publications in the database of publications in 120 psychology journals.

Not all researchers conduct quantitative research and report test statistics in their result sections. Therefore, the analysis is limited to 17 faculty members that had at least 100 test statistics. This criterion eliminated many faculty members who publish predominantly in neuroscience journals.

Figure 1 shows the z-curve for all 6,199 tests statistics. I use the Figure to explain how a z-curve analysis provides information about replicability and other useful meta-statistics.

1. All test-statistics are converted into absolute z-scores as a common metric of the strength of evidence (effect size over sampling error) against the null-hypothesis (typically H0 = no effect). A z-curve plot is a histogram of absolute z-scores in the range from 0 to 6. The 710 z-scores greater than 6 are not shown because z-scores of this magnitude are extremely unlikely to occur when the null-hypothesis is true (particle physics uses z > 5 for significance). Although they are not shown, they are included in the computation of the meta-statistics.

2. Visual inspection of the histogram shows a steep drop in frequencies at z = 1.96 (solid red line) that corresponds to the standard criterion for statistical significance, p = .05 (two-tailed). This shows that published results are selected for significance. The dashed red line shows significance for p < .10, which is often used for marginal significance. Thus, there are more results that are presented as significant than the .05 criterion suggests.

3. To quantify the amount of selection bias, z-curve fits a statistical model to the distribution of statistically significant results (z > 1.96). The grey curve shows the predicted values for the observed significant results and the unobserved non-significant results. The statistically significant results (including z > 6) make up 40% of the total area under the grey curve. This is called the expected discovery rate because the results provide an estimate of the percentage of significant results that researchers actually obtain in their statistical analyses. In comparison, the percentage of significant results (including z > 6) includes 70% of the published results. This percentage is called the observed discovery rate, which is the rate of significant results in published journal articles. The difference between a 70% ODR and a 401% EDR provides an estimate of the extent of selection for significance. The difference of~ 30 percentage points is large, but one of the smallest difference for investigations of psychology departments. The upper level of the 95% confidence interval for the EDR is 51%. Thus, the discrepancy is not just random. To put this result in context, it is possible to compare it to the average for 120 psychology journals in 2010 (Schimmack, 2022). The ODR (70% vs. 72%) is similar, but the EDR is higher (40% vs. 28%). Although this difference is not statistically significant, it suggests that the typical study at Princeton has slightly more power than studies in psychology in general.

4. The z-curve model also estimates the average power of the subset of studies with significant results (p < .05, two-tailed). This estimate is called the expected replication rate (ERR) because it predicts the percentage of significant results that are expected if the same analyses were repeated in exact replication studies with the same sample sizes. The ERR of 65% suggests a fairly high replication rate. The problem is that actual replication rates are lower than the ERR predictions (about 40% Open Science Collaboration, 2015). The main reason is that it is impossible to conduct exact replication studies and that selection for significance will lead to a regression to the mean when replication studies are not exact. Thus, the ERR represents the best case scenario that is unrealistic. In contrast, the EDR represents the worst case scenario in which selection for significance does not select more powerful studies and the success rate of replication studies is not different from the success rate of original studies. To predict the success rate of actual replication studies, I am using the average of the EDR and ERR, which is called the actual replication prediction (ARP). For Princeton, the ARP is (65 + 40)/2 = 52.5%. This is somewhat higher than the currently best estimate of the success rate for actual replication studies based on the Open Science Collaboration project (~40%). Thus, research from Columbia University is expected to replicate at a slightly higher rate than studies in psychology in general.

5. The EDR can be used to estimate the risk that published results are false positives (i.e., a statistically significant result when H0 is true), using Soric’s (1989) formula for the maximum false discovery rate. An EDR of 40% implies that no more than 8% of the significant results are false positives, but the lower limit of the 95%CI of the EDR, 31%, allows for 12% false positive results. To lower the risk of a false positive result, it is possible to reduce the significance threshold to alpha = .005 (Benjamin et al., 2017). Figure 2 shows that implications of this new criterion (z = 2.8). The false positive risk is now 2% and even the upper limit of the 95% confidence interval is only 3%. Thus, without any further information readers could use this criterion to interpret results published in articles by psychology researchers at Princeton.

Some researchers have changed research practices in response to the replication crisis. It is therefore interesting to examine whether replicability of newer research has improved. To examine this question, I performed a z-curve analysis for articles published in the past five year (2016-2021).

The point estimate of the EDR increased from 40% to 61%, but due to the relatively small number of observations this change is not statistically significant. It is also problematic that z-curve plot shows a higher frequency of z-scores between 2.2 and 2.4 rather than 2.0 and 2.2. While there are many reasons for this finding, one explanation could be that some researchers use a new criterion value for selection. Rather than publishing any p-value below .05, they may only publish p-values below .02, for example. This practice would bias the z-curve estimates that assume no further selection effects once a p-value is below .05.

The next figure shows the results for an analysis that excludes z-scores between 2 and 2.2 from the analysis. The main finding is that the EDR estimate drops from 61% to 25%. As a result, the FDR estimate increases from 3% to 16%. Thus, it is too early to conclude that Princeton’s research has become notably more replicable, and I would personally continue to use alpha = .005 to reject null-hypotheses.

10 of the 17 faculty members with useful data were classified as social psychologists. The following analysis is limited to the z-scores of these 10 faculty members to examine whether social psychological research is less replicable (Open Science Collaboration, 2015).

The EDR is slightly, but not significantly, lower, but still higher than the EDR of other departments. Thus, there is no evidence to suggest that social psychology at Princeton is less replicable than research in other areas. Other areas did not have sufficient test statistics for a meaningful analysis.

There is considerable variability across individual researchers, although confidence intervals are often wide due to the smaller number of test statistics. The table below shows the meta-statistics of all faculty members that provided results for the departmental z-curve. You can see the z-curve for individual faculty member by clicking on their name.

Rank  NameARPEDRERRFDR
1Uri Hasson8283811
2Kenneth A. Norman7375712
3Jordan A. Taylor6877604
4Elke U. Weber6670633
5Tania Lombrozo6573584
6Diana I. Tamir5969505
7Yael Niv5868476
8Emily Pronin51683410
9Jonathan D. Cohen50762317
10Alin Coman4957408
11Molly J. Crockett49752317
12J. Nicole Shelton4755398
13Susan T. Fiske46702219
14Stacey Sinclair40493112
15Eldar Shafir35551431
16Deborah A. Prentice33551144
17Joel Cooper28441239

Are there maladaptive personality traits?

The concept of personality disorders has its roots in psychiatry. Wikipedia provides a clear definition of personality disorders.

Personality disorders (PD) are a class of mental disorders characterized by enduring maladaptive patterns of behavior, cognition, and inner experience, exhibited across many contexts and deviating from those accepted by the individual’s culture.

The definition of personality disorders shares many features with definitions of normal personality traits. Personality traits are enduring dispositions that produce cross-situational and cross-temporal consistency in behaviors, cognitions, and emotions. The key difference between personality disorders and personality traits is that personality disordered traits are assumed to be maladaptive, unhealthy, or deviant from societal and cultural norms.

The history of psychiatry and psychology shows how problematic it can be when a profession is allowed to define mental disorders, especially when they are defined as deviance from social norms. The fist Diagnostic Manual of the American Psychiatric Association included homosexuality as a mental disorder (wikipedia). This is now recognized as a mistake, and social progress aims to be more inclusive towards individuals who deviate from traditional cultural norms.

Progressive forces are also trying to change social norms regarding body types, skin color, and many other attributes that vary across individuals. However, some psychologists are working towards a comprehensive system of personality disorders that may create new stigmas for individuals with deviant personality traits. This is a dangerous trend that has not received enough attention. To be maladaptive, a personality trait should have clear negative effects on individuals’ health and well-being. This requires extensive validation research and careful examination of measures that could be used to diagnose personality disorders. For example, the CATP-PD project identified 33 distinct personality disorders (Simms et al., 2011), ranging from Anhedonia to Withdrawn personality disorder.

To study personality disorders, personality disorder researchers developed questionnaires that can be used to diagnose personality disorders. Studies with these measures showed that responses to items on personality disorder questionnaires are often strongly correlated with responses to items on measures of normal personality traits (Wright & Simmons, 2014). This raises concerns about the discriminant validity of measures that are designed to assess personality disorders. For example, normal measures of personality measure individual differences in trust. Some individuals are more trusting than others. Trust or distrust can be advantageous in different contexts. However, the CAT-PD includes a measure of mistrust as one of 33 personality disorders. The challenge for theories of personality disorders is to demonstrate that the mistrust scale does not merely measure normal variation in trust, but identifies maladaptive forms or levels of low trust. This leads to two statistical criteria that a valid measure of personality disorders should fulfill. First, variation in the personality disorder measure should be distinct from variation in normal personality. Second, the unique variance in the measure of personality disorders should predict symptoms of adaptation failures. As the key criterion for a mental disorder is suffering, maladaptive personality traits should predict lower well-being (e.g., internalizing or externalizing symptoms, lower life-satisfaction).

Another threat to the validity of personality disorder measures is that self-ratings are often influenced by the desirability of items. This response bias has been called halo bias, socially desirable responding, other-deception, faking, or self-enhancement. Ample evidence shows that self-ratings are influenced by halo bias. The strongest evidence for the presence of halo bias comes from multi-rater studies and studies that compare self-ratings to objective measures (Anusic et al., 2009). For example, whereas intelligence and attractiveness are practically uncorrelated, self-ratings show a positive correlation because some individuals exaggerate their attractiveness and intelligence. Halo bias also influences self-ratings of normal personality traits and well-being. As most personality disorder items are highly evaluative, it is likely that self-ratings of personality disorders are also contaminated by halo bias. Studies with self-ratings and informant ratings also suggest that self-ratings of personality disorders are distorted by response styles (Quilty, Cosentino, & Bagby, 2018). This could mean that honest responders are misdiagnosed as having a personality disorders, whereas self-enhancers are falsely diagnosed as not having a personality disorder.

To explore the validity of the CATP-PD scales as measures of personality disorders, I reanalyzed the data from Wright and Simms (2014) article. The dataset consists of ratings on the 30 facets of Costa and McCrae’s model of normal personality, the 33 scales of the CAT-PD, and another measure of personality disorders, the PID-5. The scale scores were subjected to an exploratory factor analysis with five factors. The factors were labeled Antagonism (Manipulativeness .71, Straightforwardness -70), Negative Affectivity (Anger .71, Trust -.52), Disinhibition (Irresponsibility .75, Self-Discipline -87), Detachment (Emotional Detachment .65, Positive Emotions -.59), and Psychoticism (Unusual Beliefs, .76, no strong negative loading). This model suggests that higher order factors of normal personality and personality disorders overlap. However, this model should not be taken too seriously because it has relatively low fit. There are several reasons for this low fit. First, exploratory factor analysis (EFA) often confounds substantive factors and method factors. Confirmatory factor analysis is often needed to separate method variance from substantive variance. Second, EFA cannot represent hierarchical structures in data. This is a problem because the Big Five are higher-order factors of basic personality traits called facets. it is possible that the 33 personality disorder scales are related to normal personality at both of these levels, but factor analysis assumes that all correlations are produced by shared variance with the higher-order Big Five factors. Finally, EFA does not allow for residual correlations among Big Five facets or personality disorder scales. All of these problems explain why an EFA model fails to fit the observed correlation matrix.

To provide a better test of the validity of the CATP-PD scales as measures of personality disorders, I performed a confirmatory factor analysis (CFA). CFA is a statistical tool. The name suggests that it can only be used for confirmatory analysis, but this is not true. CFA can also be used to explore models and then confirm these models in new datasets. It is not possible to use EFA for exploration because its limitations make it impossible to find a fitting model that could be subjected to a confirmatory test. As I have only one dataset, the results are exploratory and require confirmation with new data.

Exploratory Factor Analysis

Wright and Simms (2014) did not report standard fit statistics for their EFA model. Also, I limited the analysis to the (normal) personality scales and the CAT-PD scales. Thus, I ran an EFA with five factors to obtain fit indices that can be compared to those of the CFA model, chi2 (1,648) = 3993.93, CFI = .786, RMSEA = .048, SRMR = .071, AIC = 71,534.36, BIC = 73,449.10. While the RMSEA is below the standard criterion value of .06, CFI is well below the criterion value of .95. However, these criterion values are only suggestive. More important is the comparison of alternative models. A better model should have better fit to the data, especially for fit indices that reward parsimony (CFI, RMSEA, AIC, BIC).

Confirmatory Factor Analysis

The final model was constructed in several steps starting with the model for normal personality. The theoretical model assumes six independent factors, the Big Five and a halo factor. Primary loadings of the 30 facets on the Big Five were specified according to Costa and McCrae’s theory and free. Loadings on the halo factors were initially constrained to 1, but freed in the final model. Secondary loadings were added if modification indices were greater than 20 and if they were interpretable. Finally, residual correlations were added if modification indices were above 20. The final Big Five plus Halo model (B5+H) had good fit, chi2 (314) = 556.89, CFI = .951, RMSEA = .045, SRMR = .059, AIC = 54763.01, BIC = 55,354.43. This finding already shows a problem with the 5-factor EFA solution. The EFA model failed to identify a separate openness factor, even though openness is clearly present in the data.

I then explored how the 33 CAT-PD scales are related to the 36 personality predictors; that is, the 30 facets, the Big Five, and the halo factor. I always allowed for halo effects, even if they were not significant, but I only used significant (p < .01) personality predictors. After these exploratory analysis, I fitted a model with all 33 CAT-PD scales. This required modeling relationships among CAT-PD scales that are not explained by the personality predictors (residual relationships). This led me to specify a CAT-PD method factor that resembles the fifth factor in the EFA model. The final model did not have any modification indices greater than 20.

The final model had about the same number of parameters (1,651 vs. 1,648 degrees of freedom), but the CFA model had better fit, chi2 (1,651) = 3145.88, CFI = .866, RMSEA = .038, SRMR = .069, AIC = 70,514.85, BIC = 72,416.26.

The key results are summarized in Table 1. It shows the loadings of the 33 CAT-PD scales on the seven factors. The CAT-PD scales are sorted in order of their primary loadings. The excel file can be downloaded here (results).

Eight scales had their primary loading on the Halo (evaluative) factor. Some of these loadings were very strong. For example, halo explained more than 50% of the variance in callousness scores (-.83^2 = 69%). In contrast, relationships to the five factors of normal personality were relatively weak. Only five loadings exceeded a value of .3 (9% explained variance). The strongest relationship was the loading of domineering on neuroticism (.40^2 = 16%). These scales also showed no notable loadings on the CAT factor that reflects shared variance among CAT scales. Some scales had additional relationships with specific facets of normal variation in personality. Most notably, domineering was predicted by the unique variance in assertiveness, .38^2 = .14%.

It is difficult to reconcile these findings with the conceptualization of these CAT scales as measures of maladaptive personality variation. The primary loadings on the halo factor will produce correlations with measures of adaptation that also rely on self-ratings that are influenced by halo bias (e.g., life satisfaction), but it is doubtful that these scales could predict maladaptive outcomes that are measured with independent measures. Scales that are related to neuroticism are expected to show negative relationships with measures of well-being, but it is not clear that they would predict unique variance after controlling for the known effects of neuroticism, especially the depressiveness facet of neuroticism.

My predictions about the relationship of these CAT scales and measures of well-being need to be tested with actual data, but the present findings show that studies that rely exclusively on self-ratings can be biased if the influence of halo bias on self-ratings is ignored.

The next 12 scales have their primary loading on the neuroticism factor of variation in normal personality. Some of these loadings are very high. For example, neuroticism explains .80^2 = 64% of the variance in scores on the Affective Lability scale. It is well-known that high levels of neuroticism are a risk-factor for mood disorders, especially during times of high stress. It can be debated whether this makes high neuroticism a personality disorders. Alternatively, the diathesis-stress model would argue that neuroticism is a disposition that only becomes maladaptive in combination with environmental factors. However, even if neuroticism were considered a personality disorder, it is not clear whether the 12 CAT-PD scales with primary loadings on neuroticism add to our understanding of mental health problems and mood disorders. An alternative perspective is that the CAT-PD scales merely reflect different manifestations of mood disorders. This could be tested by examining how scores on these scales respond to treatment of mood disorders. It is also noteworthy that the CAT-PD failed to identify personality disorders that are related to maladaptive low levels of negative affect. Especially, the absence of negative emotions that can inhibit behaviors such as guilt or anxiety could be maladaptive.

Five CAT-PD scales had their primary loading on the extraversion factor of normal personality. Exhibitionism was the only scale with a positive loading. The loading was high and suggested that Extraversion explained 59% of the variance. Studies of extraversion and well-being typically show that higher levels of extraversion predict higher well-being. This suggests that any maladaptive effects of exhibitionism are related to the remaining unexplained variance. Thus, it is questionable that high levels of extraversion should be considered a personality disorder.

Anhedonia and Social withdrawal are related to low Extraversion. As for CAT-PD scales related to neuroticism, it is not clear that these scales add something to our understanding of personality and mental health. Introversion itself may not be considered maladaptive. Rather, introversion may be a disposition that is only maladaptive under specific circumstances. Furthermore, causality may be reversed. Treatment of mood disorders with medication increases extraversion scores.

Extraversion explains only 10% of the variance in romantic disinterest. Thus, this scale is hardly related to variation in normal personality. Moreover, it is not clear why opposite tendencies such as hypersexuality are missing.

None of the 33 CAT-PD scales have a primary loading on Openness. This is surprising because high Openness is sometimes associated with both being a genius and being detached from reality. In any case, there is no evidence to suggest that normal variation in Openness is maladaptive.

Even more surprising is the finding that none of the 33 CAT-PD scales have their primary loadings on agreeableness. The main reasons is that disagreeableness scales are strongly influenced by socially desirable responding. Thus, more effort needs to be made to measure these constructs without halo bias. Informant ratings might be useful to obtain better measures of these constructs.

Four CAT-PD scales have their primary loadings on conscientiousness. Two had positive loadings, namely Perfectionism (.56^2 = 31%) explained variance and workaholism (.51^2 = 26% explained variance). It is not clear that this justifies considering high conscientiousness a personality disorder. Rather, high conscientiousness could be a risk factor that is only maladaptive in specific environments or it could reflect disorders that express themselves in different ways for individuals with different personality traits. For example, workaholism might be a specific maladaptive way to cope with negative affect that is not different from alcoholism or other addictions.

The last four CAT-PD scales have their primary loadings on the CAT-scale specific factor. Thus, they do not show strong overlap with normal personality. The high loading for unusual experiences might tap into cognitive symptoms related to psychoticism. Self-harm shows weak loadings on all factors and it is not clear that it measures a personality trait rather than a symptom of distress.

In conclusion, these results provide little support for the hypothesis that there is a large number of personality disorders. The main link between models of normal variation in personality and personality disorder scales is neuroticism. It is well-known that high levels of neuroticism predict lower well-being and are risk-factors for mood disorders. However, the remaining variation in personality is not consistently related to proposed measures of personality disorders.

Comparing the Results to the EFA Results

The better fitting 7-factor CFA model differs notably from the 5-factor EFA model in the original publication. The EFA model identified a single factor called Antagonism. This factor blends the halo factor and the Agreeableness factor of normal personality. As shown in the CFA model, CAT-PD scales load on the halo factor, whereas less evaluative scales of normal personality reflect Agreeableness. The reason EFA missed this distinction is that there are only six agreeableness measures and some of them have rather low loadings on agreeableness. As EFA is strongly influenced by the number of indicators for a factor, EFA failed to identify the agreeableness factor and the halo factor was misinterpreted as antagonism because antagonistic traits are highly undesirable.

The second factor was called Negative Affectivity and corresponds to the Neuroticism factor in the CFA model.

The third factor was called Disinhibition and corresponds more or less to the Conscientiousness factor in the CFA model.

The fourth factor was called Detachment and corresponds to Extraversion.

The fifth factor was called Psychoticism, which can be confusing because this is also a term used by Eysenck for variation in normal personality. This factor is similar to the CAT-specific factor in the CFA. Thus, it does not represent a dimension of normal personality.

Finally, the EFA model failed to represent Openness as a distinct factor for the same reason it failed to show a distinct agreeableness factor. As Openness is not related to PD-scales, the six Openness facets were simply not enough to form a distinct factor in a model limited to five factors.

In sum, the main problem of the EFA model is that it failed to identify agreeableness and openness as dimensions of normal personality. It therefore does not represent the well-known five factor structure of normal personality. This makes it impossible to examine how variation in normal personality is related to scales that are intended to measure personality disorders. Another problem is that EFA fails to separate method variance and content variance. In short, EFA is an inferior tool to study the relationship between measures of normal personality and allegedly maladaptive traits. The CFA model proposed here can serve as a basis for future exploration of this question, ideally with multi-rater data to separate method variance from content variance.

References

Simms, L. J., Goldberg, L. R., Roberts, J. E., Watson, D., Welte, J., & Rotterman, J. H. (2011). Computerized adaptive assessment of personality disorder: Introducing the CAT–PD Project. Journal of Personality Assessment, 93, 380–389. doi:10.1080/00223891.2011.577475

 

Replicability Rankings of Psychology Departments

Introduction

Since 2011, it is an open secret that many published results in psychology journals do not replicate. The replicability of published results is particularly low in social psychology (Open Science Collaboration, 2015).

A key reason for low replicability is that researchers are rewarded for publishing as many articles as possible without concerns about the replicability of the published findings. This incentive structure is maintained by journal editors, review panels of granting agencies, and hiring and promotion committees at universities.

To change the incentive structure, I developed the Replicability Index, a blog that critically examined the replicability, credibility, and integrity of psychological science. In 2016, I created the first replicability rankings of psychology departments (Schimmack, 2016). Based on scientific criticisms of these methods, I have improved the selection process of articles to be used in departmental reviews.

1. I am using Web of Science to obtain lists of published articles from individual authors (Schimmack, 2022). This method minimizes the chance that articles that do not belong to an author are included in a replicability analysis. It also allows me to classify researchers into areas based on the frequency of publications in specialized journals. Currently, I cannot evaluate neuroscience research. So, the rankings are limited to cognitive, social, developmental, clinical, and applied psychologists.

2. I am using department’s websites to identify researchers that belong to the psychology department. This eliminates articles that are from other departments.

3. I am only using tenured, active professors. This eliminates emeritus professors from the evaluation of departments. I am not including assistant professors because the published results might negatively impact their chances to get tenure. Another reason is that they often do not have enough publications at their current university to produce meaningful results.

Like all empirical research, the present results rely on a number of assumptions and have some limitations. The main limitations are that
(a) only results that were found in an automatic search are included
(b) only results published in 120 journals are included (see list of journals)
(c) published significant results (p < .05) may not be a representative sample of all significant results
(d) point estimates are imprecise and can vary based on sampling error alone.

These limitations do not invalidate the results. Large difference in replicability estimates are likely to predict real differences in success rates of actual replication studies (Schimmack, 2022).

Department Rankings

The main results of the replicability analysis are included in this table. Detailed analyses of departments and faculty members can be found by clicking on the hyperlink of a university.

The table is sorted by the all time actual replication prediction (ARP). It is easy to sort the table by other meta-statistics.

The ERR is the expected replication rate that is estimated based on the average power of studies with significant results (p < .05).

The EDR is the expected discovery rate that is estimated based on the average power of studies before selection for significance. It is estimated using the distribution of significant p-values converted into z-scores.

Bias is the discrepancy between the observed discovery rate (i.e., the percentage of significant results in publications) and the expected discovery rate. Bias reflects the selective reporting of significant results.

The FDR is the false discovery risk. It is estimated using Soric’s formula that converts the expected discovery rate into an estimate of the maximum percentage of false positive results under the assumption that true hypothesis are tested with 100% power.

For more information about these statistics, please look for tutorials or articles on z-curve on this blog.

UniversityARP-AllERR-ALLEDR-ALLBias-AllFDR-AllARP-5YERR-5YEDR-5YBias-5YFDR-5Y  
University of Michigan55694131858.57245276
Western University54.5703929873.5777012
University of Toronto546741288566943247
Princeton University52.5654030867.5746183
University of Amsterdam50.5663535104769254115
Harvard University4869274014556842227
Yale University4865313812557040318
University Texas - Austin46.56627441455.57041248
University of British Columbia 44672147204765293413
McGill University43.56621572043.569185723
Columbia University41.5622149193961175026
New York University41622050204870264315
Stanford University4160224518586650155

2021 Replicability Report for the Psychology Department at Columbia University

Since 2011, it is an open secret that many published results in psychology journals do not replicate. The replicability of published results is particularly low in social psychology (Open Science Collaboration, 2015).

A key reason for low replicability is that researchers are rewarded for publishing as many articles as possible without concerns about the replicability of the published findings. This incentive structure is maintained by journal editors, review panels of granting agencies, and hiring and promotion committees at universities.

To change the incentive structure, I developed the Replicability Index, a blog that critically examined the replicability, credibility, and integrity of psychological science. In 2016, I created the first replicability rankings of psychology departments (Schimmack, 2016). Based on scientific criticisms of these methods, I have improved the selection process of articles to be used in departmental reviews.

1. I am using Web of Science to obtain lists of published articles from individual authors (Schimmack, 2022). This method minimizes the chance that articles that do not belong to an author are included in a replicability analysis. It also allows me to classify researchers into areas based on the frequency of publications in specialized journals. Currently, I cannot evaluate neuroscience research. So, the rankings are limited to cognitive, social, developmental, clinical, and applied psychologists.

2. I am using department’s websites to identify researchers that belong to the psychology department. This eliminates articles that are from other departments.

3. I am only using tenured, active professors. This eliminates emeritus professors from the evaluation of departments. I am not including assistant professors because the published results might negatively impact their chances to get tenure. Another reason is that they often do not have enough publications at their current university to produce meaningful results.

Like all empirical research, the present results rely on a number of assumptions and have some limitations. The main limitations are that
(a) only results that were found in an automatic search are included
(b) only results published in 120 journals are included (see list of journals)
(c) published significant results (p < .05) may not be a representative sample of all significant results
(d) point estimates are imprecise and can vary based on sampling error alone.

These limitations do not invalidate the results. Large difference in replicability estimates are likely to predict real differences in success rates of actual replication studies (Schimmack, 2022).

Columbia University

A research assistant, Dellania Segreti, used the department website to find core members of the psychology department. She found 11 professors and 2 associate professors. This makes Columbia U one of the smaller psychology departments. She used Web of Science to download references related to the authors name and initial. An r-script searched for related publications in the database of publications in 120 psychology journals.

Not all researchers conduct quantitative research and report test statistics in their result sections. Therefore, the analysis is limited to 10 faculty members that had at least 100 significant test statistics. This criterion eliminated many faculty members who publish predominantly in neuroscience journals.

Figure 1 shows the z-curve for all 7,776 tests statistics. I use the Figure to explain how a z-curve analysis provides information about replicability and other useful meta-statistics.

1. All test-statistics are converted into absolute z-scores as a common metric of the strength of evidence (effect size over sampling error) against the null-hypothesis (typically H0 = no effect). A z-curve plot is a histogram of absolute z-scores in the range from 0 to 6. The 934 z-scores greater than 6 are not shown because z-scores of this magnitude are extremely unlikely to occur when the null-hypothesis is true (particle physics uses z > 5 for significance). Although they are not shown, they are included in the computation of the meta-statistics.

2. Visual inspection of the histogram shows a steep drop in frequencies at z = 1.96 (solid red line) that corresponds to the standard criterion for statistical significance, p = .05 (two-tailed). This shows that published results are selected for significance. The dashed red line shows significance for p < .10, which is often used for marginal significance. Thus, there are more results that are presented as significant than the .05 criterion suggests.

3. To quantify the amount of selection bias, z-curve fits a statistical model to the distribution of statistically significant results (z > 1.96). The grey curve shows the predicted values for the observed significant results and the unobserved non-significant results. The statistically significant results (including z > 6) make up 21% of the total area under the grey curve. This is called the expected discovery rate because the results provide an estimate of the percentage of significant results that researchers actually obtain in their statistical analyses. In comparison, the percentage of significant results (including z > 6) includes 70% of the published results. This percentage is called the observed discovery rate, which is the rate of significant results in published journal articles. The difference between a 70% ODR and a 21% EDR provides an estimate of the extent of selection for significance. The difference of~ 50 percentage points is large, and among the largest differences of psychology departments analyzed so far. The upper level of the 95% confidence interval for the EDR is 31%. Thus, the discrepancy is not just random. To put this result in context, it is possible to compare it to the average for 120 psychology journals in 2010 (Schimmack, 2022). The ODR (70% vs. 72%) is similar, but the EDR (21% vs. 28%) is lower, although the difference is not statistically significant and could just be sampling error.

4. The z-curve model also estimates the average power of the subset of studies with significant results (p < .05, two-tailed). This estimate is called the expected replication rate (ERR) because it predicts the percentage of significant results that are expected if the same analyses were repeated in exact replication studies with the same sample sizes. The ERR of 62% suggests a fairly high replication rate. The problem is that actual replication rates are lower than the ERR predictions (about 40% Open Science Collaboration, 2015). The main reason is that it is impossible to conduct exact replication studies and that selection for significance will lead to a regression to the mean when replication studies are not exact. Thus, the ERR represents the best case scenario that is unrealistic. In contrast, the EDR represents the worst case scenario in which selection for significance does not select more powerful studies and the success rate of replication studies is not different from the success rate of original studies. The EDR of 21% is lower than the actual replication success rate of 40%. To predict the success rate of actual replication studies, I am using the average of the EDR and ERR, which is called the actual replication prediction (ARP). For Columbia, the ARP is (62 + 21)/2 = 42%. This is close to the currently best estimate of the success rate for actual replication studies based on the Open Science Collaboration project (~40%). Thus, research from Columbia University is expected to replicate at the average rate of actual replication studies.

5. The EDR can be used to estimate the risk that published results are false positives (i.e., a statistically significant result when H0 is true), using Soric’s (1989) formula for the maximum false discovery rate. An EDR of 21% implies that no more than 19% of the significant results are false positives, but the lower limit of the 95%CI of the EDR, 13%, allows for 38% false positive results. Most readers are likely to agree that this is too high. One solution to this problem is to lower the conventional criterion for statistical significance (Benjamin et al., 2017). Figure 2 shows that alpha = .005 reduces the point estimate of the FDR to 4% with an upper limit of the 95% confidence interval of 9%. Thus, without any further information readers could use this criterion to interpret results published in articles by psychology researchers at Columbia. Of course, this criterion will be inappropriate for some researchers, but the present results show that the traditional alpha criterion of .05 is also inappropriate to maintain a reasonably low probability of false positive results.

Some researchers have changed research practices in response to the replication crisis. It is therefore interesting to examine whether replicability of newer research has improved. To examine this question, I performed a z-curve analysis for articles published in the past five year (2016-2021).

The results are disappointing. The point estimate is even lower than for all year, although the difference could just be sampling error. Mostly, these results suggest that the psychology department at Columbia University has not responded to the replication crisis in psychology, despite a low replication rate that provides more room for improvement. The ARP of 39% for research published since 2016 places Columbia University at the bottom of universities analyzed so far.

Only one area had enough researchers to conduct an area-specific analysis. The social area had 6 members with useable data. The z-curve shows a slightly lower EDR than the z-curve for all 10 faculty members, although the difference is not statistically significant. The low EDR for the department is partially due to the high percentage of social faculty members with useable data.

There is considerable variability across individual researchers, although confidence intervals are often wide due to the smaller number of test statistics. The table below shows the meta-statistics of all faculty members that provided results for the departmental z-curve. You can see the z-curve for individual faculty member by clicking on their name.

Rank  NameARPEDRERRFDR
1Jonathan B. Freeman8486821
2Janet Metcalfe7780732
3Kevin N. Ochsner50663311
4Lila Davachi48752120
5Dima Amso44632516
6Niall Bolger43573012
7Geraldine A. Downey37581726
8E. Tory Higgins34531529
9Nim Tottenham34501725
10Valerie Purdie34402714

2021 Replicability Report for the Psychology Department at U Texas – Austin

Since 2011, it is an open secret that many published results in psychology journals do not replicate. The replicability of published results is particularly low in social psychology (Open Science Collaboration, 2015).

A key reason for low replicability is that researchers are rewarded for publishing as many articles as possible without concerns about the replicability of the published findings. This incentive structure is maintained by journal editors, review panels of granting agencies, and hiring and promotion committees at universities.

To change the incentive structure, I developed the Replicability Index, a blog that critically examined the replicability, credibility, and integrity of psychological science. In 2016, I created the first replicability rankings of psychology departments (Schimmack, 2016). Based on scientific criticisms of these methods, I have improved the selection process of articles to be used in departmental reviews.

1. I am using Web of Science to obtain lists of published articles from individual authors (Schimmack, 2022). This method minimizes the chance that articles that do not belong to an author are included in a replicability analysis. It also allows me to classify researchers into areas based on the frequency of publications in specialized journals. Currently, I cannot evaluate neuroscience research. So, the rankings are limited to cognitive, social, developmental, clinical, and applied psychologists.

2. I am using department’s websites to identify researchers that belong to the psychology department. This eliminates articles that are from other departments.

3. I am only using tenured, active professors. This eliminates emeritus professors from the evaluation of departments. I am not including assistant professors because the published results might negatively impact their chances to get tenure. Another reason is that they often do not have enough publications at their current university to produce meaningful results.

Like all empirical research, the present results rely on a number of assumptions and have some limitations. The main limitations are that
(a) only results that were found in an automatic search are included
(b) only results published in 120 journals are included (see list of journals)
(c) published significant results (p < .05) may not be a representative sample of all significant results
(d) point estimates are imprecise and can vary based on sampling error alone.

These limitations do not invalidate the results. Large difference in replicability estimates are likely to predict real differences in success rates of actual replication studies (Schimmack, 2022).

University of Texas – Austin

I used the department website to find core members of the psychology department. I counted 35 professors and 6 associate professors. Not all researchers conduct quantitative research and report test statistics in their result sections. I limited the analysis to 20 professors and 3 associate professors who had at least 100 significant test statistics. As noted above, this eliminated many faculty members who publish predominantly in neuroscience journals.

Figure 1 shows the z-curve for all 10,679 tests statistics in articles published by 23 faculty members. I use the Figure to explain how a z-curve analysis provides information about replicability and other useful meta-statistics.

1. All test-statistics are converted into absolute z-scores as a common metric of the strength of evidence (effect size over sampling error) against the null-hypothesis (typically H0 = no effect). A z-curve plot is a histogram of absolute z-scores in the range from 0 to 6. The 1,559 z-scores greater than 6 are not shown because z-scores of this magnitude are extremely unlikely to occur when the null-hypothesis is true (particle physics uses z > 5 for significance). Although they are not shown, they are included in the meta-statistics.

2. Visual inspection of the histogram shows a steep drop in frequencies at z = 1.96 (solid red line) that corresponds to the standard criterion for statistical significance, p = .05 (two-tailed). This shows that published results are selected for significance. The dashed red line shows significance for p < .10, which is often used for marginal significance. Thus, there are more results that are presented as significant than the .05 criterion suggests.

3. To quantify the amount of selection bias, z-curve fits a statistical model to the distribution of statistically significant results (z > 1.96). The grey curve shows the predicted values for the observed significant results and the unobserved non-significant results. The statistically significant results (including z > 6) make up 27% of the total area under the grey curve. This is called the expected discovery rate because the results provide an estimate of the percentage of significant results that researchers actually obtain in their statistical analyses. In comparison, the percentage of significant results (including z > 6) includes 69% of the published results. This percentage is called the observed discovery rate, which is the rate of significant results in published journal articles. The difference between a 71% ODR and a 27% EDR provides an estimate of the extent of selection for significance. The difference of~ 45 percentage points is fairly large. The upper level of the 95% confidence interval for the EDR is 39%. Thus, the discrepancy is not just random. To put this result in context, it is possible to compare it to the average for 120 psychology journals in 2010 (Schimmack, 2022). The ODR (71% vs. 72%) and the EDR (27% vs. 28%) are very similar to the average for 120 psychology journals.

4. The z-curve model also estimates the average power of the subset of studies with significant results (p < .05, two-tailed). This estimate is called the expected replication rate (ERR) because it predicts the percentage of significant results that are expected if the same analyses were repeated in exact replication studies with the same sample sizes. The ERR of 66% suggests a fairly high replication rate. The problem is that actual replication rates are lower than the ERR predictions (about 40% Open Science Collaboration, 2015). The main reason is that it is impossible to conduct exact replication studies and that selection for significance will lead to a regression to the mean when replication studies are not exact. Thus, the ERR represents the best case scenario that is unrealistic. In contrast, the EDR represents the worst case scenario in which selection for significance does not select more powerful studies and the success rate of replication studies is not different from the success rate of original studies. The EDR of 27% is lower than the actual replication success rate of 40%. To predict the success rate of actual replication studies, I am using the average of the EDR and ERR, which is called the actual replication prediction (ARP). For UT Austin, the ARP is (66 + 27)/2 = 47%. This is just a bit above the currently best estimate of the success rate for actual replication studies based on the Open Science Collaboration project (~40%). Thus, UT Austin results are expected to replicate at the average rate of psychological research.

5. The EDR can be used to estimate the risk that published results are false positives (i.e., a statistically significant result when H0 is true), using Soric’s (1989) formula for the maximum false discovery rate. An EDR of 22% implies that no more than 18% of the significant results are false positives, but the lower limit of the 95%CI of the EDR, 18%, allows for 31% false positive results. Most readers are likely to agree that this is too high. One solution to this problem is to lower the conventional criterion for statistical significance (Benjamin et al., 2017). Figure 2 shows that alpha = .005 reduces the point estimate of the FDR to 2% with an upper limit of the 95% confidence interval of 5%. Thus, without any further information readers could use this criterion to interpret results published in articles by psychology researchers at UT Austin. Of course, this criterion will be inappropriate for some researchers, but the present results show that the traditional alpha criterion of .05 is also inappropriate to maintain a reasonably low probability of false positive results.

Some researchers have changed research practices in response to the replication crisis. It is therefore interesting to examine whether replicability of newer research has improved. To examine this question, I performed a z-curve analysis for articles published in the past five year (2016-2021).

The results show an improvement. The EDR increased from 27% to 41%, but the confidence intervals are too wide to infer that this is a systematic change. The false discovery risk dropped to 8%, but due to the smaller sample size the upper limit of the 95% confidence interval is still 19%. Thus, it would be premature to lower the significance level at this point. notable improvement. The muted response to the replication crisis is by no means an exception. Rather, currently the exception is Stanford University that has shown the only significant increase in the EDR.

Only one area had enough researchers to conduct an area-specific analysis. The social area had 8 members with useable data. The z-curve is similar to the overall z-curve. Thus, there is no evidence that social psychology at UT Austin has lower replicability than other areas.

There is considerable variability across individual researchers, although confidence intervals are often wide due to the smaller number of test statistics. The table below shows the meta-statistics of all faculty members that provided results for the departmental z-curve. You can see the z-curve for individual faculty member by clicking on their name.

Rank   Name ARP ERR EDR FDR
1Chen Yu7477712
2Yvon Delville7176663
3K. Paige Harden6269544
4Cristine H. Legare5873427
5James W. Pennebaker5563476
6William B. Swann53753211
7Bertram Gawronski52742913
8Jessica A. Church51772417
9David M. Buss48762021
10Jasper A. J. Smits45573311
11Michael J. Telch45672317
12Hongjoo J. Lee44701824
13Cindy M. Meston44553410
14Jacqueline D. Woolley42661824
15Christopher G. Beevers41622021
16Marie H. Monfils41671627
17Samuel D. Gosling38532218
18Arthur B. Markman38591824
19David S. Yeager38482814
20Robert A. Josephs37462814
21Jennifer S. Beer36512021
22Frances A. Champagne34551334
23Marlone D. Henderson27341922