# Supplemental Material

## Academic Placement Data and Analysis: An Update with a Focus on Gender

## Carolyn Dicey Jennings, Patrice Cobb, and David W. Vinson

Methods

To investigate placement rates among philosophy doctoral graduates, we collected placement information from universities, through self-report, and through public records. Data were obtained for 4018 doctoral graduates in philosophy. To assess the current placement environment, we used data from the 1657 philosophers who obtained doctoral degrees between 2012 and 2015. All analyses were carried out using the lme4 package in R. There were 474 cases containing missing data on one or more of the assessed covariates. To avoid introducing bias or reduction of power, a multivariate imputation by chained equations (MICE) was performed. MICE, a type of multiple imputation, is a means for handling missing data. Multiple imputation was done using the MICE package.

Model development was explorative in nature, except that previous findings suggested that gender may play a role in placement rates. To assess factors that influence placement rates in the current academic environment for philosophy doctorates our outcome variable of interest was academic placement within the first two years of obtaining a doctoral degree (or before obtaining a doctoral degree). A dummy coded variable was created to represent those obtaining a permanent academic placement versus all others, including temporary academic placements, nonacademic placements, and no placement at all. Therefore, a Bernoulli distribution, a special case of the binomial distribution, was employed. Further details on the model can be found at the end of this post.

Model Development

Model 1 is illustrated by the following equation:

Where the outcome variable, a binary response variable for obtaining permanent academic placement within the first two years, or earlier for the i-th case; is the regression intercept, or an estimate for the reference group; through are partial regression coefficients. Covariates included in the model were gender (gender), year doctorate degree was granted (cohort), first-reported area of specialization category (AOS), and doctoral granting program.

It was thought that persons obtaining a doctoral degree from the same institution may be more similar to one another that to other cases. This means that cases may vary differently depending on the institution where the doctoral degree was granted. Therefore, the above model was expanded to assess this hypothesis. A multilevel model was used to account for the possible nesting within doctoral granting institutions,

Again, , the outcome variable, for the i-th case (i = 1, … ), but with the addition of j (j = 1, ... ), where nj is the doctoral granting institution. Individual level differences were modeled at level one, gender, cohort year, and AOS. Level 2 is the group level, or granting institution, also known as the between-groups level.

Thus, the intercept for each granting institution is allowed to vary randomly . This model was used in the subsequent analyses.

Tables and Figures