Beyond research, a main interest of mine is psychometric and quantitative methods. Below are a variety of projects I have worked on that used advanced or novel statistical methods.
In terms of statistical software/programming languages, I prefer R, SPSS, and Microsoft Excel, but I've also used Matlab, Mplus, and Python. I used R for all analyses below, unless otherwise noted.
Recently, I have conducted research on Tucker’s congruence coefficient, a statistic used as an index of similarity between factor solutions from two separate factor analyses. The purpose of this research is to clarify the meaning of the congruence coefficient using a profile similarity framework. We break down the congruency coefficient into its three fundamental components: saturation (mean factor loading), scatter (variance of factor loadings), and pattern similarity (correlation between factors), and show how these components are represented within the congruence coefficient formula. Understanding how each of these components contributes to factor similarity has the potential to improve researchers’ interpretation and usage of the congruence coefficient when used for exploratory factor analysis (Hartley & Furr, under review).
Latent Profile Analysis is a type of statistical analysis that is used to find subtypes or "profiles" of related cases from multivariate continuous data. In other words, LPA, when used on social/personality psychology data, reveals whether there are different "profiles" of people based on participants' responses on a variety of measures. Latent Profile Analysis is similar to Latent Class Analysis (LCA), except that the variables are continuous whereas LCA uses categorical variables. LPA and LCA are conceptually similar to Cluster Analysis, but Cluster Analysis groups individuals based on euclidean distance whereas LPA groups individuals probabilistically.
I've used LPA to explore whether people have different “lenses” of morality: whether some individuals are highly attuned to morality and care deeply about morality when forming impressions while other individuals may care little about others’ morality and therefore are not influenced by it when forming impressions. So far, LPA has revealed there are at least two profiles of individuals that differ in how they see morality in others. I haven't found a package in R that handles LPA well, so I've used MPlus for these analyses.
I've used Generalized Estimating Equations (GEE) when dealing with repeated measures data. GEE is useful for repeated measures data because it takes correlations among the measures into account when estimating the standard errors, in contrast to other approaches, such as repeated-measures ANOVA (see Liang & Zeger, 1986). I made use of GEE in a paper I wrote with my colleagues on expert-novice differences in perceiving child behavior change (Hartley, Wright, Zakriski & McCarthy, 2014). Expert and novice clinical staff provided behavioral ratings of a child target at two points in time. We used GEE to model participants' ratings of the targets behavior across time, and added expertise as a covariate into the model.
Running GEE in SPSS is easy, and most closely resembles how SPSS handles multi-level modeling and regression. However, I wrote a simple program in R using the GEEpack package. The output is easy to interpret, and I was able to write code that extracted and plotted the predicted values (see Hartley et al., 2014)
I have also used multilevel modeling (e.g., Helzer et al., under review) to analyze relationships between situations, behavior, and personality traits using electronic sampling method (ESM) data, and growth-curve modeling to analyze behavior change. R is excellent for complex multilevel modeling and producing figures that illustrate the findings, with many useful packages (e.g., multilevel,nlme,lme4, to name just a few). But I recommend SPSS for simpler multilevel models that don't require figures because SPSS can provide results quickly and efficiently.
Beyond these projects, I have experience with a variety of other statistical methods: