All posts by Dr. R

About Dr. R

Since Cohen (1962) published his famous article on statistical power in psychological journals, statistical power has not increased. The R-Index makes it possible f to distinguish studies with high power (good science) and studies with low power (bad science). Protect yourself from bad science and check the R-Index before you believe statistical results.

Statues are Falling, but Intelligence Researchers Cling to Their Racist Past

Psychology wants to be a science. Unfortunately, respect and reputations need to be earned. Just putting the name science in your department name or in the title of your journals doesn’t make you a science. A decade ago, social psychologists were shocked to find out that for years one of their colleagues had just made up data and nobody had noticed it. Then, another social psychologists proved physics wrong and claimed to have evidence of time reversed causality in a study with erotic pictures and undergraduate student. This also turned out to be a hoax. Over the past decade, psychology has tried to gain respect by doing more replication studies of classic findings (that often fail), starting to preregister studies (which medicine has implemented years ago), and in general to analyze and report their results more honestly. However, another crisis in psychology is that most measures in psychology are used without evidence that they measure what they measure. Imagine a real science where scientists first ensure that their measurement instruments work and then use them to study distant planets or microorganisms. Not so psychology. Psychologists have found a way around proper measurement called operationalism. Rather than trying to find measures for constructs, constructs are defined by the measures. What is happiness? While philosophers have tried hard to answer this questions, psychologists cannot be bothered to spend time to think about this question. Happiness is whatever your rating on a happiness self-report measure measures.

The same cheap trick has been used by intelligence researchers to make claims about human intelligence. They developed a series of tasks and performance on these tasks is used to create a score. These scores could be given a name like “score that reflects performance on a series of tasks some White men (yes, I am a White male myself) find interesting,” but then nobody would care about these scores. So, they decided to call it intelligence. If pressed to define intelligence, they usually do not have a good answer to this question, but they also don’t feel the need to give an answer because intelligence is just a term for the test. However, the choice of the term is not an accident. It is supposed to sound as if the test measures something that corresponds to the everyday term intelligence to make the test more interesting. However, it is possible that the test is not the best measure of what we normally mean by intelligence. For example, performance on intelligence tests correlates only about r = .3 with self-ratings or ratings by close friends and family members of intelligence. While there can be measurement in self-ratings, there can also be measurement error in intelligence tests. Although intelligence researchers are considered to be intelligent, they rarely consider this possibility. After all, their main objective is to use these tests and to see how they relate to other measures.

Confusing labels for tests are annoying, but hardly worth to write a long blog post about. However, some racist intelligence researchers use the label to make claims about intelligence and skin color (Lynn & Meisenberg, 2010). Moreover, the authors even use their racist preconception that dark-skinned people are less intelligence to claim that intelligence tests measure intelligence BECAUSE performance on these tests correlates with skin color.

You don’t have to be a rocket scientists to realize that this is a circular argument. Intelligence tests are valid because they confirm a racist stereotype. This is not how real science works, but this doesn’t bother intelligence researchers. The questionable article has been cited 80 times.

I only came across this nonsense because a recent article used national IQ scores to make an argument about intelligence and homicides. After concerns about the science were raised, the authors retracted their article pointing to problems in the measurement of national differences in IQ. The editor of this journal, Psychological Science, wrote an editorial with “A Call for Greater Sensitivity in the Wake of a Publication Controversy.”

Greater sensitivity also means to clean the journals of unscientific and hurtful claims that serve no scientific purpose. In this spirit, I asked the current editor of Intelligence in an email on June 15th to retract Lynn and Meisenberger’s offensive article. Today, I received the response that the journal is not going to retract the article.

Richard Haier (Emeritus, Editor in Chief) Decision Letter

This decision just shows the unwillingness among psychologists to take responsibility for a lot of bad science that is published in their journals. This is unfortunately because it shows the low motivation to change and improve psychology. It is often said that science is the most superior method to gain knowledge because science is self-correcting. However, often scientists stand in the way of correction and the process of self-correction is best measured in decades or centuries. Max Plank famously observed that scientific self-correction often requires the demise of the old guard. However, it is also important not to hire new scientists who continue to abuse the freedom and resources awarded to scientists to spread racist ideology. Meanwhile, it is best to be careful and to distrust any claims about group differences in intelligence because intelligence researchers are not willing to clean up their act.

A Hierarchical Factor Analysis of Openness to Experience

In this blog post I report the results of a hierarchical factor analysis of 16 primary openness to experience factors. The data were obtained and made public by Christensen, Cotter, and Silvia (2019). The dataset contains correlations for 138 openness items taken from four different Big Five measures (NEO-PI3; HEXACO, BFAS, & Woo). The sample size was N = 802.

The authors used network analysis to examine the relationship among the items. In the network graph, the authors identified 10 clusters (communities) of items. Some of these clusters combine overlapping constructs in different questionnaires. For example, aesthetic appreciation is represented in all four questionnaires.

This is a good first step, but Figure 1 leaves many questions unanswered. Mainly, it does not provide quantitative information about the relationship of the clusters to each other. The main reason is that network analysis does not have a representation of the forces that bind items within a cluster together. This information was presented in a traditional correlation table based on sum scores of items. The problem with sum scores is that correlations between sum scores can be distorted by secondary loadings. Moreover, there is no formal test that 10 clusters provide an accurate representation of item-relationships. As a result, there is no test of this model against other plausible models. The advantage of structural equation modeling with latent variables is that it is possible to represent unobserved constructs like Openness and to test the fit of a model to the data.

Despite the advantages of structural equation modeling (SEM), many researchers are reluctant to use structural equation modeling for a number of unfortunate reasons. First, structural equation modeling has been called Confirmatory Factor Analysis (CFA). This has led to the misperception that SEM can only be used to test theoretical models. However, it is not clear how one would derive a theoretical that perfectly fits data without exploration. I use SEM to explore the structure of openness without an a priori theoretical model. This is no more exploratory than visual inspection of a network representation of a correlation matrix. There is no good term for this use of SEM because the term exploratory factor analysis is used for a different mathematical model. So, I simply call it SEM.

Another reason why SEM may not be used is that model fit can show that a specified model does not fit the data. It can be time consuming and require thought to create a model that actually fits the data. In contrast, EFA and network models always provide a solution even if the solution is suboptimal. This makes SEM harder to use than other exploratory methods. However, with some openness to new ideas and persistence, it is also always possible to find a fitting model with SEM. This does not mean it is the correct model, but it is also possible to compare models to each other with fit indices.

SEM is a very flexible tool and its capabilities have often not been fully recognized. While higher-order or two-level models are fairly common, models with more than two levels are rare, but can be easily fit to data that have a hierarchical structure. This is a useful feature of SEM because theoretical models have postulated that personality is hierarchically structured with several levels: The global level, aspects, facets, and even more specific traits called nuances below facets. However, nobody has attempted to fit a hierarchical model to see whether Openness has an aspect, a facet, and a nuance level. Christensen et al.’s data seemed ideally suited to examine this question.

One limitation of SEM is that modeling becomes increasingly more difficult as the number of items increases. On the other hand, three items per construct are sufficient to create a measurement model at the lowest level in the hierarchy. I therefore first conducted simple CFA analysis of items belong to the same scale and retained items with high loadings on the primary factor and no notable residual correlations with other items. I did not use the 20 aspect items because they were not designed to measure clean facets of Openness. This way, I only need to fit a total of 48 items for the 16 primary scales of Openness in the three questionnaires:

NEO: Artistic, Ideas, Fantasy, Feeling, Active, Values
HEXACO: Artistic, Inquisitive, Creative, Unconventional
Woo: Artistic, Culture, Tolerance, Creative, Depth, Intellect

Exploratory analysis showed that the creative scales in the HEXACO and Woo measures did not have unique variance and could be represented by a single primary factor. This was also the case for the artistic construct in the HEXACO and Woo measures. However, the NEO artistic items showed some unique variance and were modeled as a distinct construct, although this could just be some systematic method variance in the NEO items.

The final model (MPLUS syntax) had reasonably good fit to the data, RMSEA = .042, CFI = .903. This fit was obtained after exploratory analyses of the data and simply shows that it was possible to find a model that fits the data. A truly confirmatory test would require new data and fit is expected to decrease because the model may have overfitted the data. To obtain good model fit it was necessary to include secondary loadings of items. Cross-validation can be used to confirm that these secondary loadings are robust. All of this is not particularly important because the model is exploratory and provides a first attempt at fitting a hierarchical factor model to the Openness domain.

In Figure 2, the boxes represent primary factors that represent the shared variance among three items. The first noteworthy different to the network model is that there are 14 primary constructs compared to 10 clusters in the network model. However, Neo-Artistic (N-Artistic) is strongly related to the W/H-Artistic factor and could be combined while allowing some systematic measurement error in the NEO-items. So, conceptually, there are only 13 distinct constructs. This still leaves three more constructs than the network analysis identified. The reason for this discrepancy is that there is no strict criterion at which point a cluster may reflect to related sub-clusters.

Figure 2 shows a hierarchy with four levels. For example, creativity (W/H-Creative) is linked to Openness through an unmeasured facet (Facet-2) and artistic (W/H-Artistic). This also means that creative is only weakly linked to Openness as the indirect path is the product of the three links, .9 * .7 * .5 = .3. This means that Openness explains only 9% of the variance in the creativity factor.

In factor analysis it is common to use loadings greater than .6 for markers that can be used to measure a construct and to interpret its meaning. I highlighted constructs that are related .6 or higher with the Openness factor. The most notable marker is the NEO-Ideas factor with a direct loading of .9. This suggests that the core feature of Openness is to be open to new ideas. Another marker is Woo’s curiosity factor and mediated by the facet-2 factor, the HEXACO inquisitive factor. So, core features of Openness are being open to new ideas, being curious, and inquisitive. Although these labels sound very similar, the actual constructs are not redundant. The other indicators that meet the .6 threshold are artistic and unconventional.

Other primary factors differ greatly in their relatedness to the Openness factor. Openness to Feeling’s relationship is particularly weak, .4 * .4 = .16, and suggests that openness to feelings is not a feature of Openness or that the NEO-Feelings items are poor measures of this construct.

Finally, it is noteworthy that the model provides no support for the Big Five Aspects Model that postulates a level with two factors between Openness and Openness Factors. It is particularly troubling that the intellect aspect is most strongly related to Woo’s intellectual efficiency factor (W-Intellect, effect size r = .6), and only weakly related to the ideas factor (N-Ideas, r = .2), and the curiosity factor (W-Curious, r = .2). As Figure 2 shows, (self-rated) intellectual abilities are a distinct facet and not a broader aspect with several subordinate facets. The Openness facet is most strongly related to artistic (W/H artistic, r = .4), with weaker relationships to feelings, fantasy, and ideas (all r = .2). The problem with the development of the Big Five Aspects Model was that it relied on Exploratory Factor Analysis that is unable to test hierarchical structures in data. Future research on hierarchical structures of personality should use Hierarchical Factor Analysis.

In conclusion, SEM is capable of fitting hierarchical models to data. It is therefore ideally suited to test hierarchical models of personality. Why is nobody doing this. Orthodoxy has delegated SEM to confirmatory analysis of models that never fit the data because we need to explore before we can build theories. It requires high openness to new ideas, being unconventional, and curiosity, and inquisitiveness to break with conventions and to use SEM as a flexible and powerful statistical tool for data exploration.

Open SOEP: Spousal Similarity in Personality

Abstract: I examined spousal similarity in personality using 4-waves of data over a 12-year period in the German Socio-Economic Panel. There is very little spousal similarity in actual personality traits like the Big Five. However, there is a high similarity in the halo rating bias between spouses.

Spousal similarity in personality is an interesting topic for several reasons. First, there are conflicting folk ideas about spousal similarity. One saying assumes that “birds of the same feather flock together;” another says that “opposites attract.” Second, there is large interest in the characteristics people find attractive in a mate. Do extraverts find other extraverts more attractive? Would assertive (low agreeableness) individuals prefer a mate who is as assertive as they are or rather somebody who is submissive (high agreeableness)? Third, we might wonder whether spouses become more similar to each other over time. Finally, twin studies of heritability make the assumption that mating is random; an assumption that can be questionable.

Given so many reasons to study spousal similarity in personality, it is surprising how little attention this topic has received. A literature search retrieved only a few articles with few citations: Watson, Beer, McDade-Montez (2014) [20 citations], Humbad, Donnellan, Iacono McGue, & Burt (2010) [30 citations], Rammstedt & Schupp (2008) [25 citations]. One possible explanation for this lack of interest could be that spouses are not similar in personality traits. It is well-known that psychology has a bias against null-results; that is, the lack of statistical relationships. Another possibility is that spousal similarity is small and difficult to detect in small convenience samples that are typical in psychology. In support of the latter explanation, two of the three studies had large samples and did report spousal similarity in personality.

Humbad et al. (2010) found rather small correlations between husbands’ and wives’ personality scores in a sample of 1,296 married couples. With the exception of traditionalism, r = .49, all correlations were below r = .2, and the median correlation was r = .11. They also found that spousal similarity did not change over time, suggesting that the little similarity there is can be attributed to assortative mating (marrying somebody with similar traits).

Rammstedt and Schupp (2008) used data from the German Socio-Economic Panel (SOEP), an annual survey of representative household samples. In 2005, the SOEP included for the first time a short 15-item measure of the Big Five personality traits. The sample included 6,909 couples. This study produced several correlations greater than r = .2, for agreeableness, r = .25, conscientiousness, r = .31, and openness, r = .33. The lowest correlation was obtained for extraversion, r = .10. A cross-sectional analysis with length of marriage showed that spousal similarity was higher for couples who were married longer. For example, spousal similarity for openness increased from r = .26 for newlyweds (less than 5 years of marriage) to r = .47 for couples married more than 40 years.

A decade later it is possible to build on Rammstedt and Schupp’s results because the SOEP has collected three more waves with personality assessments in 2009, 2013, and 2017. This makes it possible to examine spousal similarity over time and to separate spousal similarity in stable dispositions (traits) and in deviations from the typical level (states).

I start with simple correlations, separately for each of the four waves using all couples that were available at a specific wave. The most notable observation is that the correlations do not increase over time. In fact, they even show a slight trend to decrease. This provides strong evidence that spouses are not becoming more similar to each other over time. An introvert who marries an extravert does not become more extraverted as a result or vice versa.

TraitW1 (N = 6263)W2 (N = 5905)W3 (N = 5404)W4 (N = 7805)

I repeated the analysis using only couples who stayed together and participated in all four waves. The sample size for this analysis was N = 1,860.


The correlations were not stronger and did not increase over time.

The next analysis examined correlations over time. If spousal similarity is driven by assortment on some stable trait, husbands’ scores in 2005 should still be correlated with wives’ scores in 2017 and vice versa. To ensure comparability for different time lags, I only used couples who stayed in the survey for all four waves (N = 1,860).

Trait2005 Trait2009 Trait2013 Trait2017 Trait
2005 Neuroticism.
2005 Extraversion.040-.02-.02
2005 Openness.
2005 Agreeableness.
2005 Conscientiousness.

The results show more similarity on the same occasion (2005/2005) than across time. Across-time correlations are all below .2 and are decreasing. However, there are some small correlations of r = .1 for Openness, Agreeableness, and Conscientiousness, suggesting some spousal similarity in the stable trait variance. Another question is why spouses show similarity in the changing state variance.

There are two possible explanations for spousal similarity in personality state variance. One explanation is that spouses’ personality really changes in sync, just like their well-being changes in the same direction over time (Schimmack & Lucas, 2010). Another explanation is that spouses’ self-ratings are influenced by rating biases and that these rating biases are correlated (Anusic et al., 2009). To test these alternative hypotheses, I fitted a measurement model to the Big Five scales that distinguishes halo bias in personality ratings from actual variance in personality. I did this for the first and the last wave (2005, 2017) to separate similarity in the stable trait variance from similarity in state variance.

The key finding is that there is high spousal similarity in halo bias. Some couples are more likely to exaggerate their positive qualities than others. After removing this bias, there is relatively little spousal similarity for the actual trait variance.

FactorTraitState 2005State 2017

In conclusion, spouses are not very similar in their personality traits. This may explain why this topic has received so little attention in the scientific literature. Null-results are often considered uninteresting. However, these findings do raise some questions. Why don’t extraverts marry extraverts or why don’t conscientious people not marry conscientious people. Wouldn’t they be happier with somebody who is similar in their personality? Research with the SOEP data suggests that that is also not the case. Maybe the Big Five traits are not as important for marital satisfaction as we think. Maybe other traits are more important. Clearly, human mating is not random, but it is also not based on matching personality traits.

We don’t forget and until Bargh apologizes we will not forgive

John Bargh is a controversial social scientists with a knack of getting significant results when others cannot (Bargh in Bartlett, 2012). When somebody failed to replicate his most famous elderly-priming results (he published two exact replication studies, 2a and 2b, that were both successful, p < .05), he wrote a blog post. The blog-post blew up in his face and he removed it. For a while, it looked as if this historic document was lost, but it has been shared online. Here is another link to it : Nothing in their heads

Personality x Situation Interactions: A Cautionary Note

Abstract: No robust and reliable interaction effects of the Big Five personality traits and unemployment on life-satisfaction in the German Socio-Economic Panel.

With the exception of late Walter Mischel, Lee Ross, and Dick Nisbett, we are all interactionists (ok, maybe Costa & Mcrae are guilty of dispositionism). As Lewin told every body in 1934, behaviour is a function of the person and the situation, and the a priori probability that the interaction effect between the two is zero (the nil-hypothesis is false) is pretty much zero. So, our journals should be filled with examples of personality x situation interactions. Right? But they are not. Every once in a while when I try to update my lecture notes and look for good examples of a personality x situation interaction I can’t find good examples. One reason is of course the long history of studying situations and traits separately. However, experience sampling studies emerged in the 1980s and the data are ideally suited to look for interaction effects. Another problem is that interaction effects can be difficult to demonstrate because you need large samples to get significant results.

This time I had a solution to my problems. I have access to the German Socio-Economic Panel (SOEP) data. The SOEP has a large sample (N > 10,000), measured the Big Five four times over a 12-year period and many measures of situations like marriage, child birth, or unemployment. So, I could just run an analysis and find a personality x situation interaction. After all, in large samples, you always get p < .05. Right? If you think so, you might be interested to read on and find out what happened.

The Big Five were measure the first time in 2005 (wave v). I picked unemployment and neuroticism as predictors because it is well-known that neuroticism is a personality predictor of life-satisfaction and unemployment is a situational predictor of life-satisfaction. It also made sense that neurotic people might respond more strongly to a negative life-event. However, contrary to these expectations, the interaction was far from significant (p = .5), while the main effects of unemployment (-1.5) and neuroticism (-.5) were highly significant. The effect of unemployment is equivalent to a change by three standard deviations in neuroticism.

Undeterred, I looked for interactions with the other Big Five dimensions. Surely, I would find an explanation for the interaction when I found one. To make things simple, I added all five interactions to the model and, hooray, a significant interaction with conscientiousness popped up, p = .02.

Was I the first to discover this? I quickly checked for articles and of course somebody else had beat me to the punch. There it was. In 2010, Boyce, Wood, and Brown had used the SOEP data to show that conscientious people respond more strongly to the loss of a job.

Five years later, a follow-up article came to the same conclusion.

A bit skeptical of p-values that are just below .02, I examined whether the interaction effect can be replicated. I ran the same analysis as I did with the 2005 data in 2009.

The effect size was cut in half and the p-value was no longer significant, p = .25. However, the results did replicate that none of the other four Big Five dimensions moderated the effect of unemployment.

So, what about the 2013 wave? Again not significant, although the effect size is again negative.

And what happened in 2017? A significant effect, hooray again, but this time the effect is positive.

Maybe the analyses are just not powerful enough. To increase power, we can include prior life-satisfaction as a predictor variable to control for some of the stable trait variance in life-satisfaction judgments. We are now only trying to predict changes in life-satisfaction in response to unemployment. In addition, we can include prior unemployment to make sure that the effect of unemployment is not due to some stable third variable.

We see that it is current unemployment that has a negative effect on life-satisfaction. Prior unemployment actually has a positive effect, suggesting some adaptation to long-term unemployment. Most important, the interaction between conscientiousness and current unemployment is not significant, p = .68.

The interaction was also non-significant in 2013, p = .69.

And there was no significant interaction in 2017, p = .38.

I am sure that I am not the first to look at this, especially given two published articles that reported a significant interaction. However, I suspect that nobody thought about sharing these results because the norm in psychology is still to report significant results. However, the key finding here appears to be that the Big Five traits do not systematically interact with a situation in explaining an important outcome.

So, I am still looking for a good demonstration of a personality x situation interaction that I can use for my lecture in the fall. Meanwhile, I know better than to use the published studies as an example.

Open Letter about Systemic Racism to the Editor of SPPS

Dear Margo Monteith,

it is very disappointing that you are not willing to retract an openly racist article that was published in your journal Social Psychological and Personality Science (SPPS) when Simine Varzire was editor of the journal and Lee Jussim was the action editor of the article in question (Cesario, Johnson, & Terrill, 2019). I have repeatedly pleaded with you to retract the article that draws conclusions on the basis of false assumptions. I am even more stunned by your decision because you rejected my commentary on this racist article with the justification that a better criticism was submitted. This criticism was just published (Ross et al., 2020). It makes the same observation that I made in my critique; that is, the conclusion that there is no racial bias in policing and the use of force rests entirely on an invalid assumption. The original authors simply assume that police officers only encounter violent criminals or that they only encounter violent criminals when they use deadly force.

Maybe you are not watching the news, but the Black Lives Matter movement started because police often use deadly force against non-violent African Americans. In some cases, this is even documented on video. Please watch the murder of Tamir Rice, George Floyd, Philando Castile, and Eric Garner and then tell their families and friends that police only kills violent criminals. That is what SPPS is telling everybody with the mantel of scientific truth, but is a blatantly false claim based on racists assumptions. So, why are you not retracting this offensive article?

Philando Castile:

Tamir Rice:

Eric Garner:

George Floyd:

So, why are you not retracting an article that makes an obviously false and offensive assumption? Do you think that a retraction would look badly on the reputation of your journal? In that case, you are mistaken. Research shows that journals that retract articles with false conclusions have higher impact factors and are more prestigious than journals that try to maintain a flawless image by avoiding retractions of bad science (Nature). So, your actions are not only offensive, but also hurt the reputation of SPPS and ultimately our science.

Your justification for not retracting the article is unconvincing.

Just how to analyze data such as this is debated, mostly in criminology journals. (One can wonder what psychology was present in Cesario et al.’s study that led to publication in SPPS, but that’s another matter.) Cesario et al. made the important point that benchmarking with population data is problematic. Their methodology was imperfect. Ross et al. made important improvements. If one is interested in this question of police bias with benchmarking, the papers bring successive advances. ”

Your response implies that you did not fully understand Ross et al.’s criticism of the offensive article. The whole approach of “benchmarking” is flawed. So, publishing an article that introduces a flawed statistical approach from criminology to psychology is dangerous. What if we would start using this approach to study other disparities? Ross et al. show that this would be extremely harmful to psychological science. It is important to retract an article that introduces this flawed statistical approach to psychologists. As an editor it is your responsibility to ensure that this does not happen.

It is particular shocking and beyond comprehension that you resist retraction at the very same time many universities and academics are keenly aware of the systemic racism in academia. This article about an issue that affects every African American was based on research funding to White academics, reviewed by White academics, approved by White academics, and now defended and not retracted by a White academic. How does your action promote diversity and inclusion? It is even more surprising that you seem to be blind to this systemic racism in the publication of this racist article given your research on prejudice and the funding you received to study these issues (CV). Can you at least acknowledge that it is very offensive to Black people to attribute their losses of lives entirely to violent crime?

Ulrich Schimmack

SPPS needs to retract Cesario’s False Claims about Racial Bias in Police Shootings

Academia is very slow in correcting itself. This is typically not a problem in psychological science because many articles do not have immediate real world consequences. However, when they do, it is important to correct mistakes as quickly as possible. The question whether (If there is any doubt about it) or how much racial bias in policing contributes to the racial disparity in victims of lethal use of force is one of them. While millions of Americans are demonstrating in the streets to support the Black Lives Matter movement, academics are slow to act and to show support for racial equality.

In 2019, the journal Social Psychological and Personality Science (SPPS) published an article by Cesario et al. with the controversial claim that there is no evidence that racial bias contributes to racial disparities in lethal use of force. The article even came to the opposite conclusion that police offers have a bias to shoot more White people than Black people. The article was edited and approved for publication by Lee Jussim, who is know for tirades against liberal-bias in academia. I cannot speak for him and he has repeatedly denied an opportunity to explain his decision. So, I have no evidence to disprove the hypothesis that he accepted the article because the conclusion fitted his conservative anti-anti-racism world-view. This would explain why he overlooked glaring mistakes in the article.

The main problem with this article is that it is unscientific. It is actually one of the worst articles I have ever seen and trust me, I have read and critiqued a lot of bad science. Don’t take my word for it. Aside from myself, SPPS received two other independent criticism of the article. My critique was rejected with the argument that one of the other criticisms was superior. After reading it, I agreed. It is a meticulous, scientific take-down of the garbage that Lee Jussim accepted for publication. I was happy that others agreed with me and made the point more clearly than I could. I was waiting patiently for it to be published. Then George Floyd was murdered on camera and the issue of racial bias in policing led to massive protests and swift actions.

During this time everybody was looking for the science on racial bias in policing. I know because my blog-posts about Cesarios’s fake science received a lot of views. The problem was that Cesario’s crappy science was published in prestigious, peer-reviewed journals, which made him the White expert on racial bias in policing. He happily responded to interview requests and presented his work as telling the true scientific story. The take down of his SPPS article that undercut his racist narrative was still not published.

On May 29, I emailed the current editor of SPPS to ask when the critique would be published.

“Dear. Dr. Monteith,    given recent events, I am wondering where we are with the response to the SPPS article that makes false claims about lethal use of force against Black Americans. Is there a preprint of the response or anything that can be shared in public? “

Margo Monteith emailed me that there is no problem with sharing the article.

“I don’t see a problem with Cody putting his article online; SAGE has agreed that it will be an open access article (and they will feature on the SPPS website). I am only posting the main points to honor the request not to publish the entire article. “

I was waiting for it to be published by SPPS, but it is still not published, so I [edited on 6/19/20] shared it on June 17. It actually was published today on June 19th (pdf). ] Everybody needs to know that there is no scientific credibility to Ceario’s claims.

However, publishing a correction is not enough. Cesario and racists ideologists like Heather MacDonald will continue to use the published articles to make false claims in public. We cannot allow this. The critic of Cesario’s article is strong enough to show that the conclusions rest entirely on racists assumptions. In short, Cesario et al simply assume that police only kill violent criminals to end up with their conclusion that given crime rates, police are too soft on violent Black criminals. The problem with this racist conclusion is clear. The assumption that police only use lethal force against known violent criminals is plain wrong and we have many videos of innocent Black victims killed by police to prove it. If you draw conclusions from a false premise, your conclusions are false. It is as simple as that. The assumption is nothing but a racist stereotype about Black people. This racist assumption should never have been published in a scientific journal. The only way to rectify the mistake is to retract the article so that Cesario can no longer use the mantel of science to spread racist stereotypes about African Americans.

Please read the rebuttal (sorry, it is a bit statistics heavy, but you can get the main points without the formulas). If you agree that the original article is flawed, I ask you to show your support with BLM and your commitment to racial equality and let SPPS know that you think the original article needs to be retracted.

Systemic Racism at Michigan State University

This is how three professors at MSU talk about innocent Black people being killed by police (podcast transcript at 25minuts and 40seconds into the clip).

Their discussion of tragic deaths suggests that Black lives don’t matter to Joseph Cesario (MSU), Steve Hsu (MSU), and Corey Washington (MSU)

Here is what those rare events look like. I dear everybody to watch them and then reflect on the words of these privileged professors.

Philando Castile:

Tamir Rice:

Eric Garner:

George Floyd:

And yes, it doesn’t only happen to Black people, but contrary to the statistically flawed work by Cesario, Young Black unarmed men are more often the target of police brutality and the victims of lethal force errors (

See also:

No Justice, No Peace: A History of Slavery Predicts Violence Today

Some human behaviors attract more attention than others. Homicides are rare, but very salient human behaviors. Governments investigate and keep records of homicides and social scientists have developed theories of homicides.

In the 1960s, social scientists suggested that inequality can lead to more violence. One simple reason is that the rewards for poor people to commit violent crimes increase with greater inequality in wealth (Becker, 1968).

Cross-national studies confirm that societies with more income inequality have higher homicide rates (Avison & Loring, 1986; Blau & Blau, 1982; Chamlin & Cochran, 2006; Corcoran & Stark, 2020; Fajnzylber, Lederman & Loayza, 2002; Krahn et al., 1986; Pratt & Godsey, 2003; Pridemore, 2008).

A recent article in Psychological Science replicated this finding (Clark, Winegard, Beardslee, Baumeister, & Shariff, 2020). However, the main focus of the article was on personality attributes as a predictor of violence. The authors main claim was that religious people are less likely to commit crimes and that among non-religious individuals those with lower intelligence would be more likely to commit homicides.

A fundamental problem with this article is that the authors relied on an article by a known White-supremacist, Richard Lynn, to measure national differences in intelligence (Lynn & Meisenberg, 2010). This article with the title “National IQs calculated and validated for 108 nations” claims that the values used by Clark et al. (2020) do reflect actual differences in intelligence. The problem is that the article contains no evidence to support this claim. In fact, the authors reveal their racist ideology when they claim that a correlation between their scores and skin color of r = -.9 validates their measure as a measure of intelligence. This is not how scientific validation works. This is how racists abuse science to justify their racist ideology.

The article also makes a common mistake to impose a preferred causal interpretation on a correlation. Lynn and Meisenberg (2010) find that their scores correlate nearly perfectly with educational attainment. They interpret this as evidence that intelligence causes educational attainment and totally ignore the plausible alternative explanation that education influences performance on logical problems. This has important implications for Clark et al.’s (2020) article because the authors buy into Lynn and Meisenberg’s racist interpretation of the correlation between performance on logic problems and educational attainment. An alternative interpretation of their finding would be that religion interacts with education. In nations with low levels of formal education, religion provides a moral code that prevents homicides. In countries with more education, other forms of ethics can take the place of religion. High levels of homicides would be observed in countries where neither religion nor education teach a moral code.

Aside from this fundamental flaw in Clark et al.’s (2020) article, closer inspection of their data shows that they overlooked confounding factors and that their critical interaction is no longer significant when these factors are included in the regression model. In fact, financial and racial inequality are much better predictors of national differences in violence than religion and the questionable measure of intelligence. Below I present the statistical results that support this conclusion that invalidate Clark et al’s (2020) racist conclusions.

Statistical Analysis

Distribution Problems

Not long ago, religion was a part of life in most countries. Only over the past century, some countries became more secular. Even today, most countries are very religious. Figure 1 shows the distribution of religiosity based on the Relig_ARDA variable in Clark et al.’s dataset. This skewed distribution can create problems when a variable is used in a regression model, especially if the variable is multiplied with another variable to test interaction effects.

It is common practice to transform variables to create a more desirable distribution for the purpose of statistical analysis. To do so, I reversed the item to measure atheism and then log-transformed the variable. To include countries that scored 100% on religiosity, I added 0.001 to all atheism scores before I carried out the log transformation. The distribution of log-atheism is less skewed.

The distribution of homicides (rates per 100,000 inhabitants) is also skewed.

Because homicide rates are right-skewed, a direct log-transformation can be applied to get a more desirable distribution. To include nations with a value of 1, I added a value of 1 before the log-transformation. The resulting distribution for log-homicides is more desirable.

The controversial IQ variable did not require a transformation.

Bivariate Relationships

The next figure shows a plot of homicides as a function of the questionable intelligence (QIM). There is a visible negative correlation. However, the plot also highlights countries in Latin America and the United States. These countries have in common that they were established by decimating the indigenous population and bringing slaves from Africa to work for the European colonialists. It is notable that nations with a history of slavery have higher homicide rates than other nations. Thus, aside from economic inequality, racial inequality may be another factor that contributes to violence even though slavery ended over 100 years ago, while racial inequality persists until today. Former slave countries also tend to score lower on the QIM measure. Thus, slavery may partially account for the correlation between QIM and homicide rates.

The next plot shows homicide rates as a function of atheism. A value of 0 would mean the country it totally atheistic, while more negative values show increasing levels of religion. There is no strong relationship between religion and homicide rates. This replicates the results in the original article by Clark et al. Remember that their key finding was a interaction between QIM and religion. However, the plot also shows a clear distinction between less religious countries. Former slave countries are low in religion and have high homicide rates, while other countries (mainly in Europe) are low in religion and have low homicide rates.

Regression Models

To examine the unique contribution of different variables to the prediction of homicide rates, I conducted several regression analyses. I started with the QIM x religion interaction to see whether the interaction is robust to transformations of the predictor variables. The results clearly show the interaction and main effects for QIM and religion (t-values > 2 are significant at p < .05).

Next I added slavery as a predictor variable.

The interaction is no longer significant. This shows that the interaction emerged because former slave countries tend to score low on QIM and religion.

I then added the GINI coefficient, the most widely used measure of income inequality, to the model. Income inequality was an additional predictor. The QIM x religion interaction remained non-significant.

I then added GDP to the model. Countries wealth is strongly related to many positive indicators. Given the skewed distribution, I used log-GDP as a predictor, which is also the most common way economists use GDP.

GDP is another significant predictor, while the QIM x religion interaction remains non-significant. Meanwhile, the strong relationship between QIM and homicide rates has decreased from b = -.71 without controls to b = -.25 with controls. However, it is still significant. As noted earlier, QIM may reflect education and Clark et al. (2020) included a measure of educational attainment in their dataset. It correlates r = .68 with QIM. I therefore substituted QIM with education.

However, education did not predict homicide rates. Thus, QIM scores capture something about nations that the education measure does not capture.

We can compare the social justice variables (slavery, GDP, GINI) with the personal-attribute (atheist, QIM) variables. A model with the social justice variables explains 62% of the variation in homicide rates across nations.

The personal-attribute model explains only 40% of the variance.

As these predictors overlap, the personal-attributes add only 3% additional variance to the variance that is explained by slavery, income inequality, and wealth.

Replicating Slavery’s Effect in the United States

The United States provide another opportunity to test the hypothesis that a legacy of slavery and racial inequality is associated with higher levels of homicides. I downloaded statistics about homicides (homicide stats). In addition, I used a measure of urbanization to predict homicides (urbanization). I also added a measure of income inequality (GINI). I classified states that fought for the confederacy as slave states (civil war facts). Results were similar for different years in which homicide rates were available from 1996 to 2018. So, I used the latest data.

In a model with all predictor variables, slavery was the only significant predictor. Income inequality showed a trend, and urbanization was not a unique predictor. When urbanization was removed from the model, the effect of income inequality was a bit stronger.

Overall, these results are consistent with the cross-national data and suggest that a history of slavery and persistent racial inequality create social conditions that lead to more violence and homicides. These results are consistent with recent concerns that systemic racism contributes to killing of civilians by civilians and police officers who historically had the role to enforce racial inequality.

Meta-Science Reflections

Clark et al.’s (2020) article is flawed in numerous ways. Ideally, the authors would have the decency to retract it. The main flaw is the use of a measure with questionable validity and to never question the validity of the measure. This flaw is not unique to this article. It is a fundamental flaw that has also led to a large literature on implicit bias based on an invalid measure. The uncritical use of measures has to stop. A science without valid measures is not a science and statistical results that are obtained with invalid measures are not scientific results.

A second flaw of the article is that psychologists are trained to conduct randomized laboratory experiments. Random assignment makes it easy to interpret statistically significant results. Unless something went really wrong or sampling error produced a false result, a statistically significant result means that the experimental manipulation influenced the dependent variable. Causality is built into the design. However, things are very different when we look at naturally occurring covariation because everything is correlated with everything. Observed relationships may not be causal and they can be produced by variables that were not measured. The only way to deal with this uncertainty is to carefully test competing theories. It is also necessary to be careful in the interpretation of results. Clark et al. (2020) failed to do so and make overly strong statements based on their correlational findings.

Many scholars have argued that religion reduces violent behavior within human social groups. Here, we tested whether intelligence moderates this relationship. We hypothesized that religion would have greater utility for regulating violent behavior among societies with relatively lower average IQs than among societies with relatively more cognitively gifted citizens. Two studies supported this hypothesis

This statement would be fine if they had conducted an experiment, but of course, it is impossible to conduct an experiment to examine this question. This also means it is no longer possible to use evidence as support for a hypothesis. Correlational evidence simply cannot verify a hypothesis. It can only falsify wrong theories. Clark et al. (2020) failed to acknowledge competing theories of homicides and to test their theory against competing theories.

The last meta-scientific observation is that all conclusions in science rests on a combination of data and assumptions. When the same data lead to different conclusions, like they did here, we get insights into researchers’ assumptions. Clark et al.’s (2020) assumptions were (a) there are notable difference in intelligence between nations, (b) these differences are measured with high validity by Lynn and Weisenberg’s (2010) questionable IQ scores, and homicides are caused by internal dispositions like being an atheist with low intelligence. Given Lynn and Weisenberg’s finding that their questionable measure correlates highly with skin tone, they also implicitly share the racist assumption that dark skinned people are more violent because they are less intelligent. The present blog post shows that an entirely different story fits the data. Homicides are caused by injustice such as unfair distributions of wealth and discrimination and prejudice based on skin color. I am not saying that my interpretation of the data is correct because I am aware that alternative explanations are possible. However, I rather have a liberal/egalitarian bias than a racist bias.