The Two High-School Pillars Supporting College Science

Reported as ~8,474 students in introductory biology, chemistry and physics courses at ~63 US colleges and universities (figures from press coverage; the parent FICSS dataset is 55 institutions from a stratified random sample of 67, ~18,000 undergraduates, of whom 11,064 were in a first course). The exact subsample could not be verified against the paper, which is paywalled with no open copy.

US undergraduates in introductory college biology, chemistry and physics, reporting retrospectively on their high school coursework; FICSS survey (NSF REC-0115649).

Retrospective self-report survey plus professor-supplied course grades, analysed with multilevel regression (students within courses within institutions). Grades converted to a 100-point scale. This is an OBSERVATIONAL study of course-taking, not of course ORDER: it measures how much biology/chemistry/physics/maths a student took, not the sequence in which they took them, and Sadler has said in interview that the design was chosen precisely because 'there aren't large numbers of people who do Physics First'. The controls (in the sibling FICSS papers) are extensive — parental education, ZIP-code community SES, public vs private high school, last HS English grade, SAT-quantitative, calculus taken, last HS maths and science grades, college year — but every one of them is a proxy, and none of them addresses the core selection problem. See genetic_confound. The famous 'out-of-discipline courses don't help' claim rests on NULL results in a moderately powered observational model; 'not significant' is not 'zero', and the paper's own reported coefficients for cross-discipline courses are not published as bounded estimates.

Two things predict a college science grade: mathematics, and the SAME discipline taken in high school. Per year of high school mathematics the model gives about +1.86 grade points (on 100) in college chemistry, +1.84 in biology, +1.28 in physics; the matching high school science course gives about +1.72 (chemistry), +1.35 (biology), +1.32 (physics). Out-of-discipline high school science — e.g. high school chemistry as preparation for college physics — is NOT significant in any of the three. Read as a sequencing claim, this is a negative result for cross-disciplinary prerequisite arguments (the usual justification for physics-first, or for chemistry-before-biology): the archive's reading is that these coefficients are small (roughly a fifth of a letter grade per year of coursework), observational, and cannot bear the weight the physics-first debate puts on them. The one signal that survives is MATHEMATICS, which is the same size or larger than in-discipline science preparation in every one of the three subjects.

HIGH — this is the decisive limitation. Students choose their high school courses, and the choice is driven by prior ability, family, tracking and school offerings, all of which are heritable or selected and all of which independently predict college science grades. The authors of the companion 2001 paper say so themselves: students who skip high school physics 'are more likely to be academically stronger, with more educated parents, having previously taken calculus'. Rough magnitude of the bite: the reported per-year coefficients are ~1.3-1.9 points on a 100-point scale where the residual SD is roughly 10, i.e. d ≈ 0.13-0.19 per YEAR of coursework, and they are estimated on a sample already selected into college science. Since SAT-quantitative and prior grades are themselves partly caused by the same coursework, the specification is simultaneously over- and under-controlled. A plausible reading is that most of the maths association is ability-selection and tracking rather than a teachable prerequisite effect, and the entire in-discipline effect is smaller than the difference between two teachers. The archive treats the whole FICSS series as grade D and non-causal — which is what its own authors say in print (2001): 'the methodology for this study is considered epidemiological rather than experimental... We understand that our study cannot claim causal connections between variables and outcomes.'

Replicated within the same research programme and dataset family rather than independently: Sadler & Tai 2001 (physics), Tai, Sadler & Loehr 2005 (chemistry, 1,531 students at 12 institutions — top predictors all mathematics-related), Schwartz, Sadler, Sonnert & Tai 2009 (depth vs breadth). No independent replication on a different dataset, and no peer-reviewed methodological critique of the FICSS selection problem appears to exist (no Science Technical Comment or Letter was published in response). The adjacent selection literature is unambiguous that this design overstates course effects: Klopfenstein & Thomas 2009 (Southern Economic Journal 10.1002/j.2325-8012.2009.tb00935.x) show that failing to control the non-AP curriculum positively biases AP coefficients to the point of vanishing; Altonji 1995 (JHR 10.2307/146029) shows returns to additional academic courses are small once selection is taken seriously; and the first RANDOMIZED study of AP science (Conger, Kennedy, Long & McGhee, JHR) finds AP takers get worse grades and higher stress than controls.