Tests of whether Qst is above or below Fst lack power in many realistic scenarios involving few populations.

In 1938, Wilks posited that the correlation of a set of linear composites approaches one when the component variables are positively correlated and the weights are also positive. Simulations confirm this theorem holds up.

The AC'RE model applied to standardized PIAT Mathematics scores.

The tests of the PIAAC are largely unbiased, with one exception.

The AC'RE model can be used to obtain the same estimand as a study of twins reared apart using cousins. Here's an application and extension of that model with height in the NLSY-79 Child and Young Adult data.

An intervention changed the heritability of growth mindset. This also led to measurement non-invariance, as expected if measurement invariance means tests measure the same things in different groups.

Applying the Method of Thresholds to Canadian LSAT performance data.

A recent study of the Flynn effect provided data to assess whether the gains were on g and the extent of bias. Bias led to generally underestimated gains, but the gains were subtest-specific and not on g.

Simulations are offered to show that power analyses should not follow from small pilot studies and that meta-analyses with publication bias won't reliably capture true effects.

Supplies two functions for working with Jané's POSC package. One extends posc_comp() to arbitrary numbers of plots and the other provides a way to make a dataframe for ggplot2 from a POSC object.

Code to check the significance of birth order effects on modern social status in Gregory Clark's data coupled with an assessment of gender interactions.

Assortative mating changes heritability, but various forces causes this to asymptote over time. This means twin-based heritability estimates are incompatible with the work of Gregory Clark. There is no current resolution.

Using event-related potentials, the measurement of intelligence is possible (i.e., r ERP-g/IQ-g ≈ 1). Since influencing neural processing speed without affecting intelligence is possible, it is highly unlikely that formative models are appropriate for intelligence.

Reanalyzing data on a Polish schooling reform that made primary school more intensive to see if it accelerated the rate of g development.

Code supplement for a paper on the Hungarian Diszkalkulia Pedagogiai Vizsgalata or DPV tool.

Plotting data from the LSAT by race and sex for the years 2006-2018. Includes MCAT data from 2022-2023.

Similar to base R

New data adds to the case for the robustness of height polygenic scores over a very long period of time.

Breit et al. (2023) produced some interesting results, but their data can't be modeled with anywhere near acceptable fit so it's unusable.

Teaching matrix rules invalidates comparisons with those who aren't taught them when it comes to figural matrices.

In Western economies, it's probably not an issue, but it becomes problematic when talking looking at Latin America and Africa.

Killingsworth (2021) used an oddly scaled axis for income that made it seem like he discovered a linear relationship between income and wellbeing. He did not.

The relationship has been misrepresented by logging a single axis. The relationship is still substantial.

The levelized cost of energy is falling for intermittent generation renewables (wind, solar), but it imposes high system costs because it necessitates a second dispatchable system that must run inefficiently.

Reanalysis of Baldwin et al., since they did not use a proper publication bias correction method. Here, I used trim-and-fill.

Simple function to calculate the proportion of votes that need to be obtained to change the result of a race.

Does education affect politics? Does politics affect education? Apparently only the former is true.

Gelman & Stern (2006) is a must-read.

Assessing a possible explanation for variance in group differences across states in the USA

Quasibinomial model of the data on genetic ancestry and self-identified race from Bryc et al. (2015)

Group differences in ACT performance in the year 2022 are examined.

Applying a method for determining mean differences and variance ratios to newly-released SAT and PSAT score data from Michigan

Ideological diversity may be negatively related to how highly colleges and universities are rated.

A simple demonstration with the Cobb-Douglas model.

A simple function to wrap lm() for HC-robust standard errors 1-5.

Will et al. (2021) was severely underpowered. Negative conclusions based on its data are not justifiable.

Nichols (1970) provided a large dataset that could be used for testing Spearman's hypothesis, so I tested it.

Humphreys et al. (2022) made strong conclusions with low-quality data and weak results.

Using data from FRED, the answer is yes.

A formula is presented to compute the expected bounds of the correlation r_{xz} given knowledge about correlations r_{xy} and r_{yz}.

In a study of dizygotic, opposite-sex twins, it was argued that hormonal explanations for male twins' elevated rates of homosexuality were ruled out. They were probably not.

Richard Hanania recently gave out a survey to his blog readers. Here is some information on the data's dimensionality and an interactive TMFG plot of it.

The barplot format was not amenable to quick visual inspection, so I turned it into a line chart.

Regraphing a plot with confidence intervals rather than a prediction interval, since the latter was widely misinterpreted.

Clustering tends to reduce within-unit correlations, making unit means unreliable proxies of the correlation in the population since they are stratified more strongly than individuals are.

RAND perform a review of the evidence on the violent crime promoting effects of concealed carry but did not assess publication bias, so I did.

A brief note on how selection thresholds do not eliminate group differences.

Simple functions for simple purposes: finding out the probability of an event happening at least once, given a known probability and number of tries, and finding the number of tries given a desired probability and a known probability per try.

European countries are generally more restrictive about abortion than American states, at least in terms of how many weeks of pregnancy women are allowed to get abortions in.

When comparing groups, absolute comparisons may need to go to the extremes.

Answering the question of how different people are given some level of relatedness, heritability, shared environmentality, and rGE.

Relationship between firearm ownership and homicide rates in each U.S. state, circa 2016.

Practice on a discrimination reaction time task did not seem to reduce the general intelligence related variance in Fleishman & Hempel's study.

Shi, Pavey & Carter (2012) provided a method for computing the Bonferroni method for controlling the FWER when outcomes are correlated, allowing for strong FWER control with greater power.

Kovacs, Molenaar & Conway (2019) made claims in favor of process overlap theory based on analyses that required more analytic leverage than their data could afford and failing to properly test for interactions.

Reanalysis of Bates & Gignac's (2022) study of incentive effects on IQ suggests that they interpreted the wrong quantity and were underpowered to detect their desired effect. Evidence is consistent with a causal relationship, but it remains to be tested.

Measurement invariance analyses for Bates & Gignac (2022). These are almost indeterminate due to the choice to use different forms for retesting and, presumably, a small intervention effect.

Copying and holding Nancy Pelosi's declared trades nets huge gains.

Functions to correct Cohen's d for reliability and to convert it to the point-biserial and back.

Taking the difference between model RMSEAs is an imprecise and insensitive method of using RMSEA. Here is a function for computing an improved version suggested by Savalei, Brace & Fouladi (2021).

Protzko (2022) claims to show that measurement invariance does not yield evidence that tests measure the same constructs in different groups. That conclusion was based on a failure to test that hypothesis and it was bolstered by the use of insensitive fit indices. Done properly, the opposite conclusion is supported.

Education does not affect intelligence, but instead, specific test scores or skills. The effect of education on SES is not mediated by cognitive gains or even subtest score gains. Intelligence is extremely stable over a nearly twenty-year period.

A recollection of Gwern's comments on the alleged failure of Nazi Germany's eugenics program to affect schizophrenia rates.

Includes a meta-analysis of assortative mating with respect to IQ and a brief investigation of the degree to which within-family PGS validity attenuation might be attributed to that phenomenon.

Functions for summary statistic t Tests and ANOVAs

Produces Yarkoni's (2009) plots and provides a function to determine the mean significant r at some sample size.

Power analysis for all groups in Pesta et al. (2020) and simple partial robustness checks.

No, that perception is based on a misunderstanding of Bayes' rule.

No. The analysis that claim was based on did not allow that conclusion.

A function to wrap subgroup_power() to determine minimum detectable effects.

For reproducibility, provide checksums for your data and data subsets and modifications.

Measurement models for empowerment and satisfaction were tested in minority and non-minority faculty. Differences were observed.

A reanalysis of the heritability meta-analysis by Pesta et al. (2020).

Function for a combination of dMACS and dMACS_Signed, for proper magnitudes and directions of effects.

An assessment of the degree of bias with respect to means by sex in cross-national personality assessment, using Johnson's IPIP-NEO 120.

An evaluation of the gender equality paradox for neuroticism alone, to serve as a data source for a comment on Benenson, Webb & Wrangham (2022).

I wanted a function for the signed version of dMACS, a formula to quantify the effect of bias in the loadings and intercepts of SEMs on the means of indicator variables.

An assumption-laden decision to filter based on variability explains the marginally significant interactions in Felig et al.'s paper. The results do not replicate with perceived actual temperature as DV.

An assessment of the relationship between verbal IQ and tolerance in the General Social Survey.

Ferreira, Almeida & Luiz (2013) provided a method for assessing change in ordinal variables. However, they did not provide code to actually use their method. Here is some code.

Checking whether an RDD for May 2020 is significant, plotting with that as a breakpoint.

Connor & Pesta recently provided additional data on this topic.

Illustrations of how the association between variables affect Venn diagrams.

Assesses subtest variance ratio differences by sex, with an attempt to remove confounding by the relationship between differences and variance ratios.

The WASI and the WRIT measure the same g factor.

Kaufman et al. correlated PAF-based g factors, so I redid their analysis with CFA.

Contains results for my analysis of the correspondence between the general factors from various cognitive test batteries.

I conducted several joint confirmatory factor analyses to assess whether different tests assessed the same g.

Supplementary material for a paper on Spearman's hypothesis.

Buczyłowska, Petermann & Daseking gave the WAIS-IV and an executive function battery to a sample of German adults. They measured the same g.

Wong et al. analyzed the relationship between "academic" and "social" intelligence. It appears they could not meaningfully distinguish them.

Found some additional datasets for assessing this question with JCFAs and other atypical tests.

Checking something from Thomas Richardson's (2021) new study.

Graphs for their results and a note about Korean adoption.

Reanalyzed Quiroga et al. (2019) with their implied higher-order model

It is commonly claimed that group differences in standardized test performance are explained by differences in the use of test preparation services.

More factor analyses for a friend of mine.

A friend needed some help factor analyzing something because he was used to STATA.

Just the meta-analytic result from an earlier analysis.

Campbell et al. (2008) tested invariance of the AGQ-M but forgot to go all the way. Here's an extended analysis.

Reupload of something previously up here.

Latent variables are used to reduce measurement error in socioeconomic status measures. This has not been justified.

Functions for working with one of Rosenthal & Rubin's methods.

A friend needed some code.

There's not very good evidence for this effect. I replicated Gignac's recent study on it.

In light of ongoing controversy regarding the cognitive effects of concussions among NFL players, I decided to chart their scores.

A colleague asked for a CFA.

R functions for Omri Tal's (2009, 2012) equations.

Addresses a gap in the visible literature.

Post-hoc power analysis of Cooper & Zubek (1958).

I analyzed whether the Brazilian BAFACALO battery and its associated dataset provided evidence for several common findings from western samples.

A cross-national effort to assess cognitive development in seven countries produced interesting and potentially useful results. I evaluated whether the countries in the sample could be compared.

I provided code to test common and independent pathway models with sibling data.

Decker et al. (2020) produced an incoherent case against current best practices in dimensionality assessment.

Gordon (1968) introduced the so-called partialing fallacy. This short post shows it is still worth worrying about.

Richardson (2020) claimed some support for sexual selection for arm length based on the relationship between armspan and UFC fight win percentages but his sample was selected.

Freis et al. (2020) put up a pretty strange preprint so I thought I would reanalyze it.

The meta-analysis by Duckworth et al. (2011) should be less than convincing to measurement-concerned researchers.

Replicates Ritchie, Bates & Deary (2015) and extends their analyses.

Prelude to a more comprehensive meta-analysis of twin model assumptions and their effects on estimates.

It is shown that the WISC-R and K-ABC model the same general factor and, despite different observed score gaps, differences in g are similarly-sized.

I've provided a function for using Naaman's "almost-sure" p-values.

Millsap proposed quantifying the proportion of the variance in observed variances which differed between groups due to differences in residual variances. I applied that to five recent parameter estimates.

Despite discussion, there has not been a comprehensive meta-analysis of the Jensen effect. I aimed to remedy that.

Gygi, Fux & Grob found bias in a model with g; reanalysis disconfirms that.

Reanalysis of the matrix and path model from an old study.

There are few published empirical estimates, so I derived some with three different methods.

Assessed interpretive errors and inconsistencies in a study by Cockroft et al. (2015).

Reanalysis of Scrivner's (2020) scale validation CFA data.