Graduate Student - Sociology and Statistics - University of Pennsylvania
In order for administrators and policymakers to make good decisions regarding fixed-term faculty, it is imperitive to establish a clear conceptualization of who these faculty are and what precisely their jobs involve. Several researchers have devised useful typologies. Tuckman (1978), Leslie and Gappa (1993), and Hollenshead et. al. (2007) have all made valuable contributions.
However, insighful quantitiative approaches may be extremely important in identifying subtle patterns and tendencies in employment. They may also be useful in identifying which dimensions of adjunct labor are more (and less) important towards this end. In the analysis below, I use machine learning (Breiman 1996) to aid in the classification of faculty. This analysis utilizes data from the Survey of Doctorate Recipients (SDR) 2013. A limitation of this dataset is that the sample only includes PhD recipients in STEM fields and is therefore not representative of non-tenure track faculty as a whole. Nonetheless, these data should be informative of an important subset of the non-tenure track population.
Using the technique of random forests, the figure to the left depicts which variables of academic labor are most important in distinguishing between those who have tenure (or are on track) and those who are tenure ineligible. Here, variable importance is defined as how much the accuracy of the procedure decreases when a particular variable of interest is excluded. In our case, the model classifies 6.5% worse when it does not draw on the important annual salary variable. A range of job and institional characteristics were considered in this procedure, including: public/private status, institution size, region, job activities, supervisory capacity, benefits, training and more. According to the model, the most important variables are salary, years in position, most common job activity and institution type.
These important variables were crosstabulated against tenure status, thereby informing us of how specifically these two faculty statuses are distributed across the most important of variables.
Our model shows that, far and away, salary is the most important factor distinguishing faculty types. Tenure/track faculty earn approximately $35,000 more per year in the STEM disciplines. Following salary, the second most important characteristic for classifying faculty is the number of years he or she has held his or her position. Tenured/track faculty on average had worked in their position over a decade (mean = 13 years). That is more than twice as long as the average duration of a NTT faculty member (mean = 6 years).
The third most important variable in this model is the job duty occupying most of faculty members' time. In this STEM dataset, 48% of tenured faculty explained that most of their time was dedicated to teaching. Only 30% of NTT faculty stated that teaching occupied most of their time. In comparison to tenured colleagues, NTT faculty spent more of their time doing research.
Education type, that is, whether an institution was a 2-year institution, 4-year college, medical school or
research institute, is also important for classifaction. We see that tenured faculty are predominately employed by 4-year institutions. Four-year institutions are also the primary employer of NTT faculty, but to a lesser extent than for tenured faculty. NTT faculty have a disproportionate presence in medical schools and research institions, once again calling attention to the role of research in the lives of STEM NTT faculty.
Hours worked per week is also an important predictor of faculty tenure status. From the crosstabulation, we see that tenured faculty in STEM work about seven more hours per week than NTT faculty. Part-time NTT faculty clearly bring the average working hours down and this is an important consideration when classifying faculty.
The carnegie classification of an institution (a measure of research intensity) is also important. NTT faculty with PhD's in STEM are frequently hired in research intensive (RI) instititions (56%). The link between research intensity and tenured faculty is considerably weaker. Only a third of tenured faculty work in R1 institions. The table reports several other predictors that aid in the process of forecasting tenure status, however, to a lesser extant that the aforementioned faculty characteristics.
In summary, the information I provided highlights some important considerations of the classification process going forward. Among STEM PhD's, an imporatant subgroup of NTT faculty, it is important to be mindful of the traditional markers of tenure status. These include salary and full-time status. However, it is also imporant to consider where faculty members are hired and trained. Large numbers of NTT faculty, we see, work not in teaching but in research. The majority of STEM PhD's working in postsecondary instititons work for schools that are research intensive and they are commonly conducting pure and applied research--not simply "teaching faculty" positions as some have assumed.
make no improvement or when a minimum node size criteria was reached.
Decision trees suggest new types of adjunct faculty emerging from the SDR. Remember, the participants in this same all had PhD's and they all came from STEM fields. So it is important to limit generalization to all NTT faculty. However, decision trees generated in this process call attention to well-compensated NTT faculty working in 4-year colleges and universities. Many of these faculty report that their job is somewhat related to their graduate training. About 3 percent fell into this category.
Another group consists of well-compensated teachers and basic researchers working in research intensive medical labs and other institutes. About 2% of the sample fell into this category.
While using re-sampling methods is useful for identifying frequent patterns, examining a single classification tree can supply insights into how the model is distinguishing between faculty member. Below is a single tree constructed on one sample of the SDR data. Salary, at the top of the tree, is once again the most importance splitting point when contructing this tree. Faculty earning more than $55,000 were grouped in the new branch going right. People earning less than $55,000 were groupeed on a new branch going left. Those new nodes were then recursively divided until the model could
Finally, there appears to be a segment working in the same kinds of institutions, but whose activities are only peripherally related to teaching or basic research. These faculty manage labs, sales, and do some programming and applied research.