Placement & AOS
Which areas of specialization have been correlated with higher placement into “permanent” (typically tenure-track or tenured) positions over the past few years?
“The broad finding is that those in Value Theory and History & Traditions (n=1125) have a higher proportion in permanent academic placement… than those in LEMM or Science, Logic, & Math (n=864),” according to Carolyn Dicey Jennings (UC Merced), who analyzed data gathered by Academic Philosophy Data and Analysis (APDA) on graduates from 2021 through 2026 and shared the results in a post at the APDA Blog.
Jennings breaks down the areas of specialization into “big” (more than 100 graduates over the past five years), “mid-sized” (between 20 and 100 graduates), and “small” (fewer than 20 graduates). Within each of those size categorizations, she lists the percentage of graduates with particular AOS’s who have found permanent placement.
Her findings are below. AOS’s listed in bold have a higher than average placement rate (the average across all areas is 32%).
Big
- Social/Political (37%; n=254)
- Ancient (37%; n=101)
- Ethics (36%; n=274)
- Epistemology (33%; n=165)
- Philosophy of Science (29%; n=102)
- Metaphysics (28%; n=112)
- Philosophy of Mind (25%; n=151)
Mid-Sized
- Law (60%; n=25)
- Medieval/Renaissance (52%; n=23)
- Gender/Race/Sexuality/Disability Studies (48%; n=56)
- Philosophy of Physics (40%; n=25)
- Logic (33%; n=39)
- German Philosophy/Kant (31%; n=68)
- Applied Ethics (31%; n=85)
- Modern Philosophy (31%; n=36)
- Historical Philosophy (29%; n=28)
- Continental Philosophy/Phenomenology (29%; n=56)
- Metaethics (26%; n=34)
- Philosophy of Religion (25%; n=20)
- Philosophy of Language (24%; n=70)
- Value Theory (22%; n=27)
- Cognitive Science/Psychology/Neuroscience/Linguistics (22%; n=73)
- Philosophy of Technology (20%, n=40)
Small
- Education (100%; n=2)
- American/Latin American Philosophy (100%; n=2)
- African Philosophy (67%; n=3)
- Philosophy of Action (67%; n=3)
- Philosophy of Biology/Environmental Philosophy (45%; n=11)
- 19th/20th Century Philosophy (44%; n=16)
- Metaphilosophy/Experimental (40%; n=5)
- Asian Philosophy (38%, n=8)
- Decision Theory (31%; n=16)
- Philosophy of Economics (23%; n=13)
- History of Analytic Philosophy (22%; n=9)
- Aesthetics (18%; n=11)
- Philosophy of Math (17%; n=18)
- Comparative Philosophy (15%; n=7)
- Other (0%; n=2)
Jennings adds: “A final point worth considering is that hiring trends come and go, and this analysis is not future-facing/predictive. Yet, the headline success of those in Value Theory and History & Traditions appears to be consistent across past analyses.”
I have a methodological concern here which makes me unsure how reliable the inter-subject comparisons are.
As I understand the methodology, APDA is looking at all the 2021-2026 graduates and seeing if, at this moment, they have a permanent job. “Permanent job” isn’t defined in the linked blog, but it presumably does include tenure-track jobs and doesn’t include fixed-term lectureships or postdocs. (I’m unclear if it includes indefinitely renewable teaching positions.)
This would not be a problem if most students either proceeded directly to a tenure-track job, or else remained indefinitely in short-term jobs or left the profession. But these days it is very common (at least among students I interact with) for people to get a tenure-track job only after doing a postdoc or two (some of which can be pretty prestigious and long-term.)
This means that the students in the early part of the 2021-2026 cohort end up being analyzed quite differently from those in the late part. A student who (say) graduated in 2021, spent one year on a postdoc at Michigan and two years on a postdoc at Harvard, and then got a TT job in 2024, would be listed as having got a permanent job. If the same student was five years later, they’d just have started their first postdoc.
That makes the absolute percentages somewhat hard to interpret: 32% of students in the analyzed cohort with named AOS got permanent jobs, but that means neither that a given student who graduated in that period has a 32% chance of a permanent job, nor that that student has a 32% chance of a permanent job within five years. But it also means that comparisons between AOSs are only meaningful if the fraction of students who proceed to a permanent job via one or more postdocs is constant across AOSs. I’m not at all convinced that’s true, both because availability of postdocs is AOS-dependent and because the postdoc route is more common in some countries than others and I don’t have reason to think the mix of international students is AOS-independent.
I’ll repeat a suggestion I’ve made previously when raising issues like this: it would be good to have data on who gets a job after 1, 3, and 5 years (analyzing, say, the three-year cohorts 2023-2025, 2021-2023, and 2019-2021 respectively, or perhaps just running all three analyses on the 2019-2021 category). If we are to look just at the placement thus far of the 2021-2026 cohort, it would also be useful to see postdoc data. (As far as I can see this isn’t available at APDA’s website, short of manually downloading each school’s data separately.)
Good point, David: some AOS’s might be more likely to seek temporary employment first, and so may seem artificially lower on an analysis that focuses on recent graduates. I will try to keep in mind your suggestion for future analyses.
A quick check I did for those interested: for those graduating 2021+, there is a significant (p=.04) difference in fellowships between these two groups (30% LEMM, 33% Value, 26% HT, and 43% SLM have a record of a fellowship or postdoc). However, if I look at only graduates between 2021 and 2023 (to allow time to find a permanent position for those who first take fellowships) the original finding still holds: 39% in Value or HT have a permanent position, while only 31% of LEMM or SLM do (p<.01).
Interesting – thanks for taking a look.
Interestingly, when I was looking over data on Pitt Philosophy’s placement 10 years ago, and talking with people about what that data suggested, this was something Tom Ricketts recommended.
I also agree that it’s comparing apples to oranges if, for the purposes of looking at permanent positions, we’re grouping people who’ve been five years on the market with those who are just coming out.
“Ethics” is among the highest, but “Value Theory” is among the lowest. What’s going on? Are these categories self-reported?
I think it’s just small-number statistics (though I agree that the actual categories are odd).
If you lump ethics, applied ethics, metaethics, and value theory together, you get 420 students total, with a success rate of 33%. If you then work out the 1-standard-deviation range around 33% for each of the four subgroups (i.e. taking as null hypothesis that there are no subarea-dependent differences in placement chance), you get:
Ethics: expected range 30%-36%, actual number 36%
Applied ethics: expected range 28%-38%
Metaethics: expected range 25%-41%, actual number 26%
Value theory: expected range 24%-42%, actual number 22%
So the value theory difference is barely significant even at 1 sigma.
In this case Value Theory refers to a generalist in Value Theory (similar to Philosophy of Science), rather than the overarching category. The detailed AOS is self reported, gathered from CVs, reported by placement officers, determined from dissertations, etc.
AOS: asceticism
I was interested in the statistical significances here so I spent a bit of time playing with the data. None of this is intended as a criticism of the original paper, and indeed I basically agree with its main finding; it’s a cautionary note aimed at anyone taking too seriously the individual-level results.
There are 1990 students (with declared AOS) in the dataset, of whom 644, or 32.3%, got permanent jobs. We can consider as a null hypothesis that each student has a 32.3% chance of success, uncorrelated with AOS. We can then ask, for each AOS, how likely it is that the higher or lower percentage observed occurred just by chance.
For example, philosophy of physics has 25 students in the dataset, of whom 10, or 40%, got jobs. The prediction on the null hypothesis is that only 8 would do so. But the probability, on the null hypothesis, of at least 10, or at most 6, getting jobs is 52%. So, taken individually, the slightly elevated success rate of philosophers of physics is (alas) probably just noise.
Analyzing the 38 subjects in the dataset, we can ask for each one individually: would this be 1% likely or less (p<0.01)? 5% likely or less? and so on.
Some results:
p<0.01: law
p<0.02: also gender/race/sexuality/disability studies; 2 total
p<0.05: no additional; 2 total
p<0.1: also cognitive science, philosophy of mind, medieval/renaissance; 5 total
p<0.2: also social/political, ethics, technology; 7 total
p<0.3: also language, math; 9 total.
With 38 individual subjects in the dataset, on the null hypothesis naively you’d expect (typically) none to be significant at p<0.02, 4 to be significant at p<0.1, 8 to be significant at p<0.2, 11 to be significant at p<0.3. (That’s probably not quite accurate here, because of discrete-sample effects in small subjects, but it should be qualitatively similar.)
So, that suggests that with the plausible exception of law and GRSD, no individual AOS result should be taken seriously.
However, we can get a bit more signal if we clump AOSs. Using the PGR categories, I get:
LEMM: n=542, 29% jobs, p=0.07
History: n=245, 36% jobs, p=0.34
Value: n=730, 35% jobs, p=0.12
Science: n=337, 28% jobs, p=0.07
Other: n=136, 39% jobs, p=0.12
Certainly a noticeable result here, albeit not meeting the formal threshold (p<0.05) for statistical significance.
If we remove the two outliers (which, yes, you can challenge, but this is a blog comment and not a preregistered study, and those are the two results that are above what we’d have expected to see by chance), we get
LEMM: n=542, 29% jobs, p=0.19
History: n=245, 36% jobs, p=0.23
Value (minus Law): n=705, 34% jobs, p=0.19
Science: n=337, 28% jobs, p=0.15
Other (minus GRSD): n=80, 33% jobs, p=1
Unsurprisingly this makes the signal a lot weaker .
The OP further clumps History with Value and Other, and LEMM with Science. At this point I start worrying about inadvertent data dredging, especially since I don’t think these clumps were visible in previous analyses; still, my Bayesian prior is reasonable that these have somewhat similar job markets. I get
LEMM+Science: n=879, 28% jobs, p=0.01
History+Value+Other: n=1111, 36% jobs, p=0.02
which FWIW is well above the statistical-significance threshold.
If again I exclude Law and GRSD, we get
LEMM+Science: n=879, 28% jobs, p=0.07
History+Value+Other: n=1030, 34% jobs, p=0.1
– no longer formally statistically significant, but certainly suggestive.
Finally, if we recheck the individual AOSs against the revised null hypothesis that LEMM+Science have different % success rates than History+Value but that there is no correlation within the groups, we get:
p<0.01: Law
p<0.02: no extras
p<0.05: also GRSD
p<0.1: also medieval / renaissance
p<0.2: also technology
p<0.3: also social/political, mind, math
No real change here: again, Law and GRSD look genuinely high and the rest looks like noise.
In summary, I think the data:
(1) fairly strongly support Law and GRSD having unusually good placement rates;
(2) support, though somewhat weakly, the hypothesis that there is a placement rate difference between LEMM+Science and History+Value+Other even once Law and GRSD are set aside
(3) don’t support any conclusions about specific AOS other than those entailed by (1)-(2).
The granularity of AOS is an interesting thing to consider since it does seem most job postings treat AOS as somewhat fungible. Perhaps that fungibility is not translating directly to hireability, but I can imagine that LEMM is more comparable to Ethics than Metaphysics is to Ethics.
Hi David. Thanks for running these numbers. Your point about small-sample noise and the pitfalls of multiple hypothesis testing is a helpful corrective to over-interpreting the individual percentages. However, treating job placements as independent events might miss some of the structural constraints of the academic market. If we apply models that account for non-independence, the conclusions shift in a few important ways.
First, the total number of permanent academic jobs in a given period is relatively fixed, making this closer to a zero-sum game. In statistical terms, this resembles sampling without replacement from a finite population. If we apply a finite population correction to the variance, the margin for random fluctuation shrinks. Under this assumption, the gap in placement rates between the LEMM/Science cluster and the History/Value cluster becomes far more pronounced. Rather than being a marginally significant fluctuation, the data points to a systemic divide.
Second, academic hiring is rarely area-neutral. Job ads usually target specific subfields, meaning candidates are mostly competing within isolated tracks rather than a single general pool. What looks like random statistical noise in the smaller AOS categories might actually reflect concrete supply-and-demand mismatches. For example, the high demand for undergraduate service courses in ethics or political philosophy naturally generates more job lines than advanced logic. The varying placement rates likely measure this structural quota rather than pure chance.
Finally, candidates are not independent data points; their outcomes are nested within institutional reputations and advisor networks. In a mixed-effects model, we would treat the graduating department as a random effect to account for this clustering. High intra-class correlation, where students from the same top program all tend to get hired, drastically reduces the effective sample size. This means the high placement rates for Law and GRSD—which survived your independence tests—might not actually stem from the AOS itself. If those graduates are heavily concentrated in a handful of elite, well-resourced programs, their apparent statistical advantage could drop significantly once institutional clustering is controlled for.
Approaching the data through these systemic constraints suggests that the underlying mechanics of the market shape these outcomes much more than random variation alone. Thanks again for prompting such an interesting look at the numbers.