Thursday, May 20, 2010
Over-Interpreted Data 3
This example of data that does not support the inferences drawn from it is quite subtle. In the Campaign for Science and Engineering in the UK Policy Document Number 8, May 2008, ‘Delivering Diversity: Making Science & Engineering Accessible to All’ is the graph shown in Figure 1. It is noted that more women entered the system as researchers and were promoted at every subsequent level. However, the comment is also made that, while in an equal world the graph would be flat so that the percentage of women at higher levels matched that at lower levels, there is no sign of the rate of increasing under-representation abating, i.e. the slope in 2005/06 is pretty much the same as the slope in 1995/96. But the graph shows that the percentage of women at each level in 2005/06 matches that of the next lowest level ten years previously, which is roughly what you would expect as each cohort moves through the system. Since the percentage of women among researchers rose for the slope to flatten the percentage of women among professors, senior lecturers and lecturers would have to have risen faster. This raises the question: is the fall-off in the percentage of women with grade the result of current discrimination or disadvantage preventing women from moving through the system from researcher to professor or is it the result of discrimination or disadvantage that occurred before 1995. If the latter it is hard to see what we can do about it now.
In fact the profile in 1995/96 can be evolved to the profile in 2005/06 without making the assumption that women are disadvantaged in appointments or promotions in that the number of women appointed as lecturers or promoted to senior lecturer or professor matches the number available in the pool from which they are appointed or promoted. The results of a model in which the percentage of women among researchers rises linearly from 24% to 30% and the percentage of women among appointments and promotions matches the percentage in the pool are shown in Figure 2. The results from the simulation for 2005 match the data rather well. (A detailed description of the model I used to produce the simulated 2005/06 figures is available at Google Docs.) This does not prove that women did not experience discrimination or disadvantage in appointments or promotion during the period 1995 to 2005. What it does show is that the data in Figure 1 do not, by themselves, show whether they did or they did not. Of course, discrimination or disadvantage must have occurred before 1995 in order to have a starting profile that was skewed to lower grades in the first place, but I was not aware that there was any dispute about that.
In the US the National Academies Report: Gender Differences at Critical Transitions in the Careers of Science, Engineering, and Mathematics Faculty (National Academies Press, 2009) found that there were gender differences in hiring. Women were more likely to receive the first job offer than they were to be asked to interview and more likely to be interviewed but the proportion of women applying was smaller than the proportion among those receiving Ph.D. degrees. Women were also more likely to receive tenure but the proportion of women candidates for tenure was smaller than the proportion of women among assistant professors. Because the data are snapshot data it was not possible to distinguish between the possibility that women are more likely to leave before coming-up for tenure and the possibility that the numbers of women being hired as assistant professors had increased in recent years. There were no gender differences in promotion to full professor. Studies of this kind that examine the transition processes with large enough data sets to reveal differences are what is required to reveal whether and when women are disadvantaged.