This repository accompanies the article *Improved goodness of fit procedures for confirmatory factor analysis* submitted to Psychological Methods We present a data base of eigenvalues and fit statistics extracted from models fitted to simulated datasets. Note that this data only contains correct model specification, so it can not be used to investigate power. The simulation has the following factors: - Seven sample sizes: 200 400 800 1500 3000 10000 1e+05 - Three model sizes: 10, 20, or 40 observed variables - Seven distributions: norm, vale-maurelli, independent generator, and piecewise linear. The non-normal distributions come in two types: moderate and severe marginal nonnormality Hence there are 441 000 datasets in the database. For each simulated dataset the database contains the following information: - p-values for 36 robustified test statistics - the test statistics ML and RLS, together with p-values - All eigenvalues $\lambda$ of $U\Gamma$ for two estimators of $\Gamma$ + First, the asymptotically unbiased estimator $\hat{\Gamma}_\text{A}$. This is the default in most software packages + Second, the finite-sample unbiased estimator $\hat{\Gamma}_\text{U}$ The database object is a list with two elements: - The first element, named "res", has length 441000. Each element in "res" has the following information: + p-values for 36 robustified test statistics + the test statistics ML and RLS, together with p-values + The eigenvalues $\lambda$ of $U\Gamma$ for two estimators of $\Gamma$ - First, the asymptotically unbiased estimator $\hat{\Gamma}_\text{A}$. This is the default in most software packages - Second, the finite-sample unbiased estimator $\hat{\Gamma}_\text{U}$ - The second element, named "p" gives the simulation conditions. It is a data frame with 441 000 rows and four variables: + n sample size + seed for random number generator + dim is the number of observed variables in model + dist is the distribution