Corrigendum to “Academic Achievement and Economic Attainment: Reexamining Associations Between Test Scores and Long-Run Earnings”

Abstract

Watts, T. W. (2020). Academic achievement and economic attainment: Reexamining associations between test scores and long-run earnings. AERA Open, 6(2). https://doi.org/10.1177/2332858420928985

On working with the data for this project after the publication of the article, the author discovered several small coding errors involving control variables used in the analysis. In general, these errors involved responses for “I don’t know” or “NA” receiving numeric codes on several variables. None of the primary measures used in the analysis (i.e., test scores or earnings) were affected by this error. The family income variable for age 16 was also recoded because approximately 16% of cases had family income undercounted due to inconsistencies in missing data values between those reporting monthly versus weekly income (updated descriptives for age-16 family income are shown below). These coding updates led to almost no change in the key results.

The updated data-cleaning syntax has been added to the OpenICSPR page corresponding to this article, along with an additional README explaining which variables were altered (https://www.openicpsr.org/openicpsr/project/119183/). Please consult this updated README before using the posted syntax files for analytic purposes. It is recommended that anyone using the syntax files for this analysis use the updated cleaning file for data preparation. The multiple imputation and analysis files have not been altered.

For sake of clarity, the key results tables from the main text of the article have been reproduced here using the updated data preparation file with the recoded control variables. As these tables reflect, alterations to the control variables led to very minor changes in the results—with most regression coefficients altered at only the third decimal place. These minor changes do not affect the findings or conclusions from the published article. Changes to tables shown in the supplemental material for the article are similarly minor, and the updated supplemental material is available in the online version of the original article.

The corrected Tables 2 to 7 are provided below:

Table 2

High School Descriptive Characteristics by Average Adult Earnings

	Men			Women
	Full sample	Top 50% earners	Bottom 50% earners	Full sample	Top 50% earners	Bottom 50% earners
Mathematics (0 to 31)	14.25	16.79	11.96	12.64	15.08	10.61
Reading (0 to 35)	26.29	28.57	24.24	26.38	28.54	24.58
Socioemotional and personality ratings (scaled 1 to 5)
Impulsive to cautious	3.26	3.22	3.30	3.25	3.23	3.26
Lazy to hardworking	3.20	3.42	3.00	3.50	3.72	3.32
Withdrawn to sociable	3.60	3.75	3.46	3.74	3.82	3.68
Rigid to flexible	3.24	3.31	3.18	3.29	3.36	3.24
Moody to even-tempered	3.62	3.72	3.53	3.58	3.70	3.48
Timid to aggressive	2.95	3.00	2.90	2.90	2.92	2.88
Demographics
Financial problems in past year (prop.)	0.09	0.06	0.11	0.09	0.06	0.11
Family monthly income (2018 USD)	2943	3114	2788	2930	3071	2813
Family owns home (prop.)	0.52	0.61	0.44	0.52	0.61	0.44
School characteristics
School enrollment	922.56	915.48	928.94	921.51	913.90	927.86
Student/teacher ratio	16.81	16.61	17.00	16.98	16.74	17.18
Percent of boys in school taking 0-levels	27.60	34.20	21.66	16.24	20.82	12.42
Percent of girls in school taking 0-levels	18.03	22.20	14.27	26.82	34.08	20.76
Observations	3826	1812	2014	2772	1262	1510

Note. The top 50% of earners represent men or women that, on average, were at or above the 50th percentile in earnings across each of the earnings waves (sample restricted to participants who indicated full-time work at a given follow-up wave). The behavioral and personality scales were measured on a continuum from “1” to “5” (e.g., for “Timid to Aggressive” a value of “1” would indicate maximum timidness and a value of “5” would indicate maximum aggressiveness). Prop. = proportion.

Table 3

Associations Between High School Test Scores and Log-Monthly Earnings Conditional on Working Full Time for Men

	Bivariate	Math and reading only	Family background and health	Behaviors and personality	Cognitive ability and pretests	Concurrent characteristics
	A	B	C	D	E	F
	Age 33 years
	(1)	(2)	(3)	(4)	(5)	(6)
Mathematics	0.130*** (0.007)	0.079*** (0.009)	0.066*** (0.009)	0.065*** (0.009)	0.043*** (0.012)	0.040*** (0.012)
Reading	0.141*** (0.007)	0.086*** (0.009)	0.075*** (0.009)	0.066*** (0.009)	0.054*** (0.012)	0.051*** (0.012)
p-value of difference						.525
N	2,847
	Age 41 years
	(7)	(8)	(9)	(10)	(11)	(12)
Mathematics	0.159*** (0.010)	0.115*** (0.014)	0.101*** (0.014)	0.103*** (0.014)	0.091*** (0.019)	0.082*** (0.019)
Reading	0.157*** (0.010)	0.075*** (0.014)	0.068*** (0.014)	0.058*** (0.015)	0.048** (0.018)	0.042* (0.018)
p-value of difference						.145
N	2,838
	Age 46 years
	(13)	(14)	(15)	(16)	(17)	(18)
Mathematics	0.164*** (0.009)	0.122*** (0.012)	0.106*** (0.013)	0.108*** (0.013)	0.086*** (0.016)	0.075*** (0.016)
Reading	0.168*** (0.011)	0.077*** (0.013)	0.072*** (0.014)	0.064*** (0.014)	0.055** (0.018)	0.043^* (0.017)
p-value of difference						.212
N	2,325
	Age 50 years
	(19)	(20)	(21)	(22)	(23)	(24)
Mathematics	0.167*** (0.011)	0.118*** (0.014)	0.102*** (0.015)	0.106*** (0.015)	0.088*** (0.019)	0.082*** (0.019)
Reading	0.171*** (0.014)	0.085*** (0.019)	0.081*** (0.020)	0.076*** (0.022)	0.056* (0.022)	0.057* (0.022)
p-value of difference						.456
N	2,205

Note. Robust standard errors are presented in parentheses. Test scores were transformed to z scores. The “p-value of difference” rows list p values from post hoc tests that tested whether the math and reading coefficients were equal to one another. The first row of the table lists the additional control variables added in the models presented in each column (e.g., for the models listed in column C, behavioral and personality measures were added to the already included set of family background and health controls). For the full list of control variables included in each column, see Supplementary Table A1.

p < .05. **p < .01. ***p < .001.

Table 4

Associations Between High School Test Scores and Log-Monthly Earnings Conditional on Working Full Time for Women

	Bivariate	Math and reading only	Family background and health	Behaviors and personality	Cognitive ability and pretests	Concurrent characteristics
	A	B	C	D	E	F
	Age 33 years
	(1)	(2)	(3)	(4)	(5)	(6)
Mathematics	0.161*** (0.010)	0.105*** (0.012)	0.094*** (0.013)	0.086*** (0.013)	0.073*** (0.016)	0.058*** (0.016)
Reading	0.187*** (0.013)	0.108*** (0.015)	0.097*** (0.017)	0.088*** (0.017)	0.080*** (0.021)	0.069** (0.021)
p-value of difference						0.688
N	1,315
	Age 41 years
	(7)	(8)	(9)	(10)	(11)	(12)
Mathematics	0.136*** (0.017)	0.078*** (0.019)	0.075*** (0.018)	0.076*** (0.018)	0.056** (0.021)	0.049^* (0.023)
Reading	0.163*** (0.017)	0.111*** (0.019)	0.102*** (0.020)	0.098*** (0.020)	0.090*** (0.026)	0.080** (0.026)
p-value of difference						0.387
N	1,670
	Age 46 years
	(13)	(14)	(15)	(16)	(17)	(18)
Mathematics	0.147*** (0.012)	0.077*** (0.016)	0.067*** (0.016)	0.064*** (0.016)	0.037 (0.021)	0.035 (0.022)
Reading	0.186*** (0.012)	0.134*** (0.016)	0.120*** (0.017)	0.116*** (0.017)	0.095*** (0.020)	0.091*** (0.021)
p-value of difference						0.092
N	1,550
	Age 50 years
	(19)	(20)	(21)	(22)	(23)	(24)
Mathematics	0.159*** (0.012)	0.093*** (0.016)	0.080*** (0.018)	0.079*** (0.018)	0.058** (0.019)	0.052** (0.020)
Reading	0.191*** (0.015)	0.126*** (0.020)	0.123*** (0.021)	0.119*** (0.022)	0.097*** (0.026)	0.089*** (0.027)
p-value of difference						0.322
N	1,600

Note. See Table 3 note.

p < .05. **p < .01. ***p < .001.

Table 5

Associations Between Composite Math and Reading Scores and Log-Monthly Earnings Conditional on Working Full Time for Men

	Bivariate	Family background and health	Behaviors and personality	Cognitive ability and pretests	Concurrent characteristics
	A	B	C	D	E
	Age 33 years
	(1)	(2)	(3)	(4)	(5)
Achievement composite	0.150*** (0.007)	0.128*** (0.008)	0.120*** (0.008)	0.088*** (0.014)	0.083*** (0.014)
N	2,847
	Age 41 years
	(6)	(7)	(8)	(9)	(10)
Achievement composite	0.175*** (0.010)	0.155*** (0.012)	0.149*** (0.012)	0.129*** (0.022)	0.114*** (0.022)
N	2,838
	Age 46 years
	(11)	(12)	(13)	(14)	(15)
Achievement composite	0.185*** (0.010)	0.165*** (0.011)	0.161*** (0.012)	0.132*** (0.020)	0.110*** (0.019)
N	2,325
	Age 50 years
	(16)	(17)	(18)	(19)	(20)
Achievement composite	0.187*** (0.012)	0.168*** (0.016)	0.168*** (0.018)	0.134*** (0.023)	0.128*** (0.023)
N	2,205

Note. See Table 3 note. The “achievement composite” variable is the standardized average of the age-16 math and reading tests.

p < .05. **p < .01. ***p < .001.

Table 6

Associations Between Composite Math and Reading Scores and Log-Monthly Earnings Conditional on Working Full Time for Women

	Bivariate	Family background and health	Behaviors and personality	Cognitive ability and pretests	Concurrent characteristics
	A	B	C	D	E
	Age 33 years
	(1)	(2)	(3)	(4)	(5)
Achievement composite	0.194*** (0.011)	0.174*** (0.013)	0.158*** (0.014)	0.139*** (0.022)	0.114*** (0.022)
N	1,315
	Age 41 years
	(6)	(7)	(8)	(9)	(10)
Achievement composite	0.169*** (0.018)	0.159*** (0.018)	0.157*** (0.019)	0.128*** (0.027)	0.113*** (0.030)
N	1,670
	Age 46 years
	(11)	(12)	(13)	(14)	(15)
Achievement composite	0.188*** (0.012)	0.166*** (0.014)	0.159*** (0.014)	0.112*** (0.025)	0.106*** (0.025)
N	1,550
	Age 50 years
	(16)	(17)	(18)	(19)	(20)
Achievement composite	0.196*** (0.013)	0.181*** (0.014)	0.176*** (0.015)	0.136*** (0.024)	0.123*** (0.025)
N	1,600

Note. See Table 5 note.

p < .05. **p < .01. ***p < .001.

Table 7

Pooled Models: Associations Between High School Test Scores and Log-Average Earnings Between Age 33 and Age 50 Conditional on Working Full Time

	Math and reading separately		Achievement composite		Achievement composite with high school controls
	Men	Women	Men	Women	Men	Women
	(1)	(2)	(3)	(4)	(5)	(6)
Mathematics	0.069*** (0.011)	0.049*** (0.013)	—	—	—	—
Reading	0.048*** (0.011)	0.082*** (0.015)	—	—	—	—
Achievement composite	—	—	0.107*** (0.013)	0.115*** (0.016)	0.101*** (0.013)	0.109*** (0.017)
Socioemotional skills and personality
Impulsive to cautious	−0.015 (0.008)	−0.011 (0.009)	−0.016^* (0.008)	−0.011 (0.009)	−0.014 (0.008)	−0.010 (0.009)
Lazy to hardworking	0.028*** (0.008)	0.018 (0.011)	0.029*** (0.008)	0.018 (0.011)	0.032*** (0.008)	0.022^* (0.011)
Withdrawn to sociable	0.041*** (0.008)	0.006 (0.009)	0.041*** (0.008)	0.006 (0.009)	0.043*** (0.008)	0.007 (0.009)
Rigid to flexible	−0.001 (0.008)	0.007 (0.009)	−0.001 (0.008)	0.007 (0.009)	−0.002 (0.008)	0.008 (0.009)
Moody to even-tempered	−0.008 (0.009)	0.000 (0.010)	−0.008 (0.009)	0.000 (0.010)	−0.009 (0.009)	−0.001 (0.010)
Timid to aggressive	0.027*** (0.008)	0.007 (0.010)	0.027*** (0.008)	0.007 (0.010)	0.027*** (0.008)	0.009 (0.010)
Control variables
Follow-up wave fixed effects	Inc.	Inc.	Inc.	Inc.	Inc.	Inc.
Fam background and health	Inc.	Inc.	Inc.	Inc.	Inc.	Inc.
Socioemotional skills and personality	Inc.	Inc.	Inc.	Inc.	Inc.	Inc.
Cognitive ability and pretests	Inc.	Inc.	Inc.	Inc.	Inc.	Inc.
Concurrent characteristics	Inc.	Inc.	Inc.	Inc.	Inc.	Inc.
Pooled observations	10,215	6,135	10,215	6,135	10,215	6,135
Unique observations	3,826	2,772	3,826	2,772	3,826	2,772
R ²	0.190	0.162	0.190	0.162	0.193	0.162
Adjusted R²	0.182	0.149	0.182	0.149	0.184	0.149

Note. Robust standard errors were adjusted for student-level clustering and are presented in parentheses. All independent variables shown were transformed to z scores. All models include the full set of controls used in the “Column F” models of Table 3. Pooled models were generated by treating each respective earnings measure (taken at ages 33, 41, 46, and 50, respectively) as an independent observation, and “follow-up” wave fixed effects were included in each regression. Columns 1 and 2 correspond to the models shown in Tables 3 and 4 (i.e., math and reading achievement modeled independently). Columns 3 and 4 correspond to the “achievement composite” estimates shown in Tables 5 and 6. In columns 5 and 6, measures indicating high school quality were added to the model. Inc. = included.

p < .05. **p < .01. ***p < .001.