Pre-1994 SAT

This consists of a variety of previously written posts on the old SAT. It will be rewritten in a cohesive way in the future.

Is the SAT an IQ test?

The SAT after 1994 is no longer an IQ test, as College Board deliberately redesigned the test to mirror high school coursework. "Thanks to an unprecedented assault from the head of the University of California system, the College Board (the nonprofit organization that owns the SAT) has begun its biggest overhaul ever of the test"¹. In early 1994, the verbal section dropped antonyms, doubled the share of passage-based reading, and the math section began allowing calculators and open-ended responses. These changes were repeated in subsequent updates to the test, diluting its saturation with the general intelligence factor (g). Due to these changes, the modern SAT moved from an aptitude test to a scholastic achievement test which can definitely be practiced for. However, this wiki will be specifically referring to the SAT forms before 1994, which have been found to be psychometrically equivalent to a Full Scale IQ test.

Directly admitted by the College Board president, Gaston Caperton, "in its original form [the SAT] was an IQ test."¹ In 2004, Frey & Detterman, using a National Longitudinal Survey of Youth subsample who had taken the old SAT, found the composite score correlated r = 0.82 with g extracted from the ten subtest ASVAB, and r = 0.72 (range-restricted) with Raven's Advanced Progressive Matrices, a well known fluid reasoning test².

Fig. 1. Scatter plots of Scholastic Assessment Test (SAT) scores and IQ estimates: first-factor score (IQ scale) from the Armed Services Vocational Aptitude Battery (ASVAB) as a function of (a) SAT total score and (b) unstandardized predicted IQ based on SAT total score, SAT2, and SAT3 and (c) Raven's Advanced Progressive Matrices score (IQ scale) as a function of SAT total score²

Furthermore, as pointed out by Frey & Detterman, "it is evident from these results that there is a striking relation between SAT scores and measures of general cognitive ability. In fact, when one examines the results in [Fig. 2.], especially those in the ASVAB column, it appears that the SAT is a better indicator of g, as defined by the first factor of the ASVAB, than are some of the more traditional intelligence tests."²

Fig. 2. Intercorrelation matrix of the SAT with other well known tests of g²

Another independent study of the SAT's value as an IQ test confirms the above findings. "In a study of 339 undergraduates, Brodnick and Ree (1995) used covariance structure modeling to examine the relationship between psychometric g, socioeconomic variables, and achievement-test scores. They found substantial general-factor loadings on both the math (.698) and the verbal (.804) SAT subtests."² While they used the SAT itself to define their first factor as g, the evidence strongly suggests it measures the same first factor g measured by IQ tests. Another thing that should be kept in mind is that these loadings are deflated due to Spearman’s Law of Diminishing Returns (SLODR), as the sample of students who took SATs were above average, college-bound high school graduates, placing them above the average 100 IQ population.

Why was the old SAT so g-loaded? Its creator, Princeton psychologist Carl Brigham, lifted item formats directly from the World War I Army Alpha intelligence tests he developed, meaning the exam's backbone was abstract analogies, antonyms, and logic puzzles that were always intended as an IQ test (and also the exact formats which the post-1994 revisions have removed).

The common objection of the SAT being skewed by the amount of prep time invested by test takers is directly contradicted by large-scale College Board studies, which put coaching gains at ~9-15 points on verbal and ~15-18 points on math³. Furthermore, there are heavy diminishing returns to the amount of time spent, as demonstrated in Fig. 3 below.

Fig. 3. The diminishing returns to coaching for the SAT⁴

The gains shown above equate to approximately between one to six IQ points- far too small to explain uncorrected correlations in the .70-.80 range with independent IQ measures. One theory as to the resistance of the SAT to the practice effect compared to other tests is its unique property of having multiple forms. Contrast that with most professionally administered IQ tests (such as the WAIS-IV or SB-V), which rely on a single copyrighted form that proctors have to guard. If a client or an internet leak reveals those items, the whole instrument is compromised until the publisher can fund and norm an alternate edition (a process which takes years). By design, the SAT's rotating forms limit any item-specific exposure that inflates retest scores on pro tests.

Few IQ tests have ever combined the accuracy of a top-tier FSIQ battery with the scale, form security, and predictive power of the pre-1994 SAT. With multiple peer reviewed independant studies reporting g correlations on par with how gold standard, professional tests correlate with each other, the old SAT can definitely stand with more conventional IQ tests. However, where the SAT is unique is, unlike pro tests given to a few thousand volunteers, the SAT was normed on millions of examinees every year and continuously equated every year. A vast, rotating item bank also meant each administration kept coaching effects trivial. Given the SAT's predictive validity with college and even mid-career outcomes in samples exceeding 200,000 students, the old SAT may be the most underappreciated intelligence test ever created.

---

National SAT Averages Converted to IQ (1955-1983)

Year	SAT-V	SAT-M	V+M
1955	98.5	103.1	100.9
1960	101.7	102.3	102.1
1966	102.7	100.6	101.8
1974	101.0	101.4	101.3
1983	101.9	102.4	102.3

These are using older norms, but they still show that this test is immune to the Flynn effect.

---

Intended Major Field	Average IQ	Mean SATV	Mean SATM	Mean SATV+SATM	Percent Planning Graduate Degree
Physics	126	558	641	1199	89
Interdis./other sci.	120	520	589	1109	77
Astronomy	120	526	578	1104	86
Economics	120	519	576	1095	81
International rel.	119	544	546	1090	82
Chemical engineering	119	490	589	1079	75
Chemistry	118	500	572	1072	78
Math & statistics	117	469	593	1062	65
Aerospace engineering	116	472	555	1027	63
Political science	115	507	515	1022	76
"Other" engineering	115	460	559	1019	65
Biological sciences	114	480	524	1004	81
Mechanical engin.	114	442	543	985	53
Electrical engin.	113	436	543	979	57
Civil engineering	113	436	533	969	51
Earth & environ. sci.	112	458	489	947	65
"Other" social sci.	110	458	467	925	61
Arch./Environ. engin.	109	419	494	913	56
General psychology	109	448	463	911	78
Computer science	109	413	489	902	46
Social psychology	108	439	451	890	67
Child psychology	106	415	428	843	72
Sociology	106	414	429	843	50
Agriculture	106	404	436	840	31
Law enforcement	103	381	408	789	33

---

Norms to Convert to IQ

Composite Norms

SAT Composite	IQ	SAT Composite	IQ
1600	166	1000	114
1590	163	990	114
1580	161	980	113
1570	159	970	113
1560	157	960	112
1550	155	950	112
1540	154	940	111
1530	153	930	110
1520	152	920	110
1510	151	910	109
1500	150	900	109
1490	149	890	108
1480	148	880	108
1470	147	870	107
1460	146	860	107
1450	145	850	106
1440	144	840	106
1430	143	830	105
1420	142	820	105
1410	141	810	104
1400	140	800	103
1390	139	790	103
1380	139	780	102
1370	138	770	101
1360	138	760	101
1350	137	750	100
1340	137	740	99
1330	136	730	99
1320	136	720	98
1310	135	710	97
1300	134	700	97
1290	133	690	96
1280	133	680	95
1270	132	670	95
1260	131	660	94
1250	130	650	93
1240	130	640	93
1230	129	630	92
1220	128	620	91
1210	127	610	90
1200	127	600	89
1190	126	590	88
1180	125	580	87
1170	124	570	86
1160	124	560	85
1150	123	550	84
1140	122	540	83
1130	122	530	82
1120	121	520	81
1110	120	510	80
1100	120	500	79
1090	119	490	77
1080	119	480	75
1070	118	470	73
1060	117	460	71
1050	117	450	69
1040	116	440	67
1030	116	430	65
1020	115	420	63
1010	115	410	61
		400	58

Subtest Norms

Subtest	Verbal Score	Math Score
800	159	152
790	156	149
780	154	147
770	152	145
760	150	143
750	148	141
740	146	139
730	144	138
720	142	137
710	141	136
700	140	135
690	139	134
680	138	133
670	137	132
660	135	131
650	134	129
640	133	128
630	132	126
620	130	125
610	129	123
600	127	122
590	126	121
580	124	120
570	123	119
560	122	118
550	121	117
540	120	116
530	119	115
520	118	114
510	117	113
500	116	112
490	114	111
480	113	110
470	112	109
460	111	108
450	110	107
440	109	106
430	108	105
420	107	104
410	106	103
400	105	101
390	103	100
380	102	99
370	101	98
360	100	97
350	99	95
340	97	94
330	96	92
320	95	91
310	93	89
300	92	87
290	91	85
280	89	83
270	87	81
260	85	78
250	83	75
240	81	72
230	79	69
220	77	66
210	75	63
200	72	59

---

Citations

Footnotes

1. https://www.mail-archive.com/futurework@scribe.uwaterloo.ca/msg05978.html ↩ ↩²

2. https://gwern.net/doc/iq/high/smpy/2004-frey.pdf ↩ ↩² ↩³ ↩⁴ ↩⁵

3. https://files.eric.ed.gov/fulltext/ED562660.pdf ↩

4. https://onlinelibrary.wiley.com/doi/epdf/10.1002/j.2333-8504.1980.tb01209.x ↩