School district investments in general skills: The case of principal residency programs

Minh Nguyen; Steven G. Rivkin; Lauren Sartain; Jeffrey C. Schiman

doi:10.2478/izajole-2023-0001

Table 1

Number of Residency-program Graduates by Cohort of Program Completion, and Number of First-year Elementary School Principals and Assistant Principals by Year and Residency-Program Participation Status

		First-year principals		First-year assistant principals
	Residency program graduates	Residency program participants	Non-participants	Residency program participants	Non-participants
2013	63
2014	65	5	36	18	70
2015	63	18	25	19	68
2016	39	18	23	17	40
2017	35	15	27	12	25
2018		14	15	6	28
Total	265	70	126	72	231

[i] Notes: First-year principal or assistant principals are counted once in the first year they appear in a CPS school. The sample of first-year principals represents those that are included in our estimation sample.

Table 2

Descriptive Statistics for the Principal Sample

	Residents	Non-residents	Difference (Residents minus non-Residents)
Student Variables
Math z-score	−0.076 [0.062]	−0.010 [0.046]	−0.065
Reading z-score	−0.075 [0.057]	−0.022 [0.038]	−0.053
Black	0.487 [0.050]	0.370 [0.040]	0.117^*
Hispanic	0.367 [0.048]	0.452 [0.038]	−0.085
Other race	0.042 [0.010]	0.069 [0.017]	−0.027
Female	0.489 [0.004]	0.497 [0.003]	−0.007
Special Education	0.150 [0.005]	0.145 [0.004]	0.005
Grade Average Enrollment	61.599 [4.269]	70.345 [3.695]	−8.746
Principal variables
Asian	0.000 [0.000]	0.014 [0.014]	−0.014
Black	0.015 [0.015]	0.010 [0.010]	0.005
Hispanic	0.425 [0.062]	0.350 [0.044]	0.075
Multiple race	0.143 [0.044]	0.167 [0.041]	−0.024
Female	0.754 [0.052]	0.587 [0.050]	0.167^**
Birth Year	1977.436 [0.717]	1971.143 [0.579]	6.293^***
Proportion with the job title: acting	0.034 [0.020]	0.049 [0.021]	−0.015
Proportion with the job title: interim	0.524 [0.064]	0.671 [0.047]	−0.147^*
Proportion with the job title: contract	0.441 [0.064]	0.280 [0.044]	0.161^**
Number of observations	23,356	41,465

Notes: Student characteristics are measured in the year prior to the entry of a the new principal or assistant principal.

Standard deviation in square brackets. Statistical significance of the resident/non-resident difference is based on regressions with standard errors clustered by school.

*** p < 0.01,

** p < 0.05,

* p < 0.1.

Table 3

Resident-Versus Non-Resident-Principal Transition Rates, by Destination and Period Examined

	Transition in any year (2014–2017)			Transition following the 1st year
Any destination
Resident Principal	0.130^* (0.068)	0.110 (0.070)	0.071 (0.083)	−0.00017 (0.050)	0.0018 (0.052)	−0.028 (0.060)
Mean Turnover	0.30	0.30	0.30	0.10	0.10	0.10
Sample Size	184	184	184	184	184	184
Multinomial Logit Estimates (relative to no transition)
Transition within CPS
Resident Principal	2.65^*** (0.77)	2.65^*** (0.77)	3.30^** (1.38)	2.15^* (1.18)	2.40^** (1.18)	2.20 (1.80)
Transition out of CPS
Resident Principal	−0.10 (0.42)	−0.16 (0.44)	−0.55 (0.60)	−1.33 (0.86)	−1.36 (0.96)	−2.39^** (1.18)
Sample Size	184	184	184	184	184	184
Student Controls	N	Y	Y	N	Y	Y
Principal Controls	N	N	Y	N	N	Y

Notes: We focus on newly entering principals between 2014 and 2017. The sample includes 184 newly entering principals. The left side of the table focuses on transitions in any year from 2014 to 2017. The right focuses on transitions occurring after the end of the new principal's first year. Student controls include school-year averages of racial composition, enrollment, share in special education, and share female. Principal controls include indicators for race, female, and job title (interim principal, acting principal, or principal) and birth year. All regressions control for year dummy variables. Robust standard errors are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.001.

Table 4

Gardner (2022) two-stage event-study estimates

		Levels		Gains
		Math	Reading	Math	Reading
Pre-period	Resident^*(t=−5)	−0.0048 (0.0155)	−0.0116 (0.0106)	0.0101 (0.0086)	0.0001 (0.0057)
Pre-period	Resident^*(t=−4)	0.0108 (0.0142)	0.0109 (0.0090)	0.0210^* (0.0108)	0.0183^** (0.0085)
Resident^*(t=−3)	−0.00004 (0.0102)	0.0061 (0.0083)	−0.0021 (0.0104)	−0.0033 (0.0074)
Resident^*(t=−2)	0.0070 (0.0131)	0.0084 (0.0091)	−0.0013 (0.0099)	0.0048 (0.0083)
Post-period	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	0.0451 (0.0305)	0.0374 (0.0240)	0.0516^*** (0.0153)	0.0321^*** (0.0122)
Resident^*(t=2)	0.0714^* (0.0427)	0.0481^* (0.0277)	0.0503^** (0.0195)	0.0165 (0.0111)
Sample Size	440968	440968	352705	351864

Notes: The samples include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2018. Schools have at least 5 years of data prior to the entry and at least 2 years following the entry of the new principal. Note that schools where a new principal entered in 2018 will only have 2 “post-period” years.

Estimates are based on the two-stage approach in Gardner (2022), where the first stage uses untreated observations to estimate school and period effects, and these effects are then removed prior to comparing outcomes for treated and untreated observations. Regressions include school fixed effects and year fixed effects. Regressions also control for indicators for grade. Period t=0 which is the first year of the new principal's tenure is dropped from the sample.

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Table 5

Gardner (2022) two-stage event-study estimates (school has 3 post-period years)

		Levels		Gains
		Math	Reading	Math	Reading
Pre-period	Resident^*(t=−5)	−0.0035 (0.0153)	−0.0031 (0.0105)	0.0046 (0.0091)	0.0009 (0.0062)
Pre-period	Resident^*(t=−4)	0.0021 (0.0129)	0.0057 (0.0089)	0.0131 (0.0091)	0.0114 (0.0073)
Resident^*(t=−3)	−0.00007 (0.0115)	0.0028 (0.0093)	0.0041 (0.0108)	−0.0006 (0.0076)
Resident^*(t=−2)	0.0072 (0.0141)	0.0053 (0.0101)	−0.0007 (0.0108)	0.0047 (0.0098)
Post-period	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	0.0417 (0.0363)	0.0502^* (0.0286)	0.0437^** (0.0183)	0.0416^*** (0.0147)
Resident^*(t=2)	0.0721^* (0.0438)	0.0494^* (0.0291)	0.0439^** (0.0197)	0.0124 (0.0116)
Sample Size	385322	385322	308109	307332

Notes: The samples include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2017. Schools have at least 5 years of data prior to the entry and 3 years following the entry of the new principal. Note that schools where a new principal entered in 2018 will not contribute to these estimates.

Estimates are based on the two-stage approach in Gardner (2022), where the first stage uses untreated observations to estimate school and period effects, and these effects are then removed prior to comparing outcomes for treated and untreated observations. Regressions include school fixed effects and year fixed effects. Regressions also control for indicators for grade. Period t=0 which is the first year of the new principal's tenure is dropped from the sample.

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Table 6

Gardner (2022) two-stage event-study estimates (principal remains there 3 years)

		Levels		Gains
		Math	Reading	Math	Reading
Pre-period	Resident^*(t=−5)	0.0076 (0.0198)	0.0011 (0.0126)	0.0117 (0.0119)	0.0005 (0.0069)
Pre-period	Resident^*(t=−4)	0.0074 (0.0157)	0.0019 (0.0098)	0.0116 (0.0106)	0.0049 (0.0067)
Resident^*(t=−3)	−0.0085 (0.0139)	0.0038 (0.0119)	−0.0011 (0.0129)	0.0053 (0.0096)
Resident^*(t=−2)	0.0071 (0.0183)	0.0031 (0.0129)	−0.0031 (0.0139)	0.0025 (0.0127)
Post-period	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	0.0466 (0.0520)	0.0616 (0.0413)	0.0519^* (0.0269)	0.0541^*** (0.0210)
Resident^*(t=2)	0.1007 (0.0660)	0.0787^* (0.0467)	0.0654^** (0.0276)	0.0247 (0.0165)
Sample Size	229918	229918	183850	183468

Notes: The samples include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2016. To be included in the sample, principals must remain in their school at least 3 years. Thus, principals from entry cohorts in 2017 and 2018 will not contribute to these estimates.

Estimates are based on the two-stage approach in Gardner (2022), where the first stage uses untreated observations to estimate school and period effects, and these effects are then removed prior to comparing outcomes for treated and untreated observations. Regressions include school fixed effects and year fixed effects. Regressions also control for indicators for grade. Period t=0 which is the first year of the new principal's tenure is dropped from the sample,

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Table 7

Gardner (2022) two-stage event-study estimates (school covariates)

		Enrollment	Black	Hispanic	Other	Female	Special Education
Pre-period	Resident^*(t=−5)	−0.2652 (0.7526)	0.0093^*** (0.0031)	−0.0054^* (0.0033)	0.0015 (0.0010)	0.0022 0.0022	0.0001 0.0019
	Resident^*(t=−4)	−0.0998 (0.4900)	0.0039^** (0.0019)	−0.0034 (0.0023)	0.0017 (0.0011)	−0.0004 0.0016	0.0004 0.0017
	Resident^*(t=−3)	−0.1110 (0.3483)	−0.0006 (0.0011)	−0.0015 (0.0011)	0.0006 (0.0007)	0.0015 0.0017	−0.0005 0.0013
	Resident^*(t=−2)	0.2523 (0.4461)	−0.0038^* (0.0022)	0.0029 (0.0023)	−0.0009 (0.0006)	0.0012 0.0018	−0.0007 0.0015
Pos	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	−1.1356 (1.2912)	−0.0155^** (0.0058)	0.0131^** (0.0066)	−0.0039 (0.0029)	0.0019 0.0045	−0.0047 0.0041
	Resident^*(t=2)	−2.1136 (1.1582)	−0.0152^* (0.0081)	0.0185^** (0.0091)	−0.0051 (0.0037)	0.0027 0.0055	−0.0052 0.0048
	Sample Size	440968	440968	440968	440968	440968	440968

Notes: Similar to Table 4, the sample here include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2018. Schools have at least 5 years of data prior to the entry and at least 2 years following the entry of the new principal. Note that schools where a new principal entered in 2018 will only have 2 “post-period” years.

Estimates are based on the two-stage approach in Gardner (2022), where the first stage uses untreated observations to estimate school and period effects, and these effects are then removed prior to comparing outcomes for treated and untreated observations. Regressions include school fixed effects and year fixed effects. Regressions also control for indicators for grade. Period t=0 which is the first year of the new principal's tenure is dropped from the sample,

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Appendix Table A1

TWFE event-study estimates

		Levels		Gains
		Math	Reading	Math	Reading
Pre-period	Resident^*(t=−5)	0.0087 (0.0431)	0.0034 (0.0274)	0.0477^* (0.0270)	0.0285 (0.0175)
Pre-period	Resident^*(t=−4)	0.0258 (0.0404)	0.0286 (0.0246)	0.0576^** (0.0284)	0.0459^** (0.0206)
Resident^*(t=−3)	0.0154 (0.0213)	0.0244 (0.0217)	0.0322 (0.0266)	0.0230 (0.0171)
Resident^*(t=−2)	0.0212 (0.0267)	0.0236 (0.0169)	0.0298 (0.0199)	0.0270 (0.0183)
Post-period	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	0.0541^** (0.0267)	0.0501^** (0.0201)	0.0779^*** (0.0215)	0.0528^*** (0.0172)
Resident^*(t=2)	0.0782^* (0.0411)	0.0584^** (0.0241)	0.0759^*** (0.0255)	0.0357^*** (0.0154)
Sample Size	440968	440968	352705	351864

Notes: The samples include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2018. Schools have at least 5 years of data prior to the entry and at least 2 years following the entry of the new principal. Note that schools where a new principal entered in 2018 will only have 2 “post-period” years.

Estimates are based on a two-way fixed effect regression with school fixed effect, year fixed effects, and grade indicators. Period t=0 which is the first year of the new principal's tenure is dropped from the sample.

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Appendix Table A2

TWFE event-study estimates (school has 3 post-period years)

		Levels		Gains
		Math	Reading	Math	Reading
Pre-period	Resident^*(t=−5)	−0.0047 (0.0434)	0.0043 (0.0303)	0.0297 (0.0268)	0.0190 (0.0202)
Pre-period	Resident^*(t=−4)	0.0028 (0.0399)	0.0151 (0.0269)	0.0386 (0.0262)	0.0304 (0.0201)
Resident^*(t=−3)	0.0019 (0.0318)	0.0129 (0.0245)	0.0282 (0.0249)	0.0176 (0.0200)
Resident^*(t=−2)	0.0090 (0.0241)	0.0148 (0.0186)	0.0207 (0.0242)	0.0216 (0.0231)
Post-period	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	0.0347 (0.0304)	0.0490^** (0.0239)	0.0606^** (0.0236)	0.0528^** (0.0212)
Resident^*(t=2)	0.0667^* (0.0392)	0.0540^** (0.0243)	0.0630^** (0.0248)	0.0284^* (0.0168)
Sample Size	385322	385322	308109	307332

Notes: The samples include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2017. Schools have at least 5 years of data prior to the entry and 3 years following the entry of the new principal. Note that schools where a new principal entered in 2018 will not contribute to these estimates.

Estimates are based on a two-way fixed effect regression with school fixed effect, year fixed effects, and grade indicators. Period t=0 which is the first year of the new principal's tenure is dropped from the sample.

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Appendix Table A3

TWFE event-study estimates (principal remains there 3 years)

		Levels		Gains
		Math	Reading	Math	Reading
Pre-period	Resident^*(t=−5)	0.0178 (0.0602)	0.0055 (0.0395)	0.0436 (0.0396)	0.0113 (0.0277)
Pre-period	Resident^*(t=−4)	0.0181 (0.0549)	0.0085 (0.0344)	0.0436 (0.0391)	0.0176 (0.0264)
Resident^*(t=−3)	0.0021 (0.0446)	0.0114 (0.0341)	0.0291 (0.0365)	0.0185 (0.0303)
Resident^*(t=−2)	0.0163 (0.0372)	0.0099 (0.0283)	0.0286 (0.0368)	0.0140 (0.0319)
Post-period	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	0.0526 (0.0426)	0.0622^* (0.0341)	0.0767^** (0.0350)	0.0614^** (0.0292)
Resident^*(t=2)	0.1004^* (0.0561)	0.0777^* (0.0394)	0.0861^** (0.0331)	0.0331 (0.0246)
Sample Size	229918	229918	183850	183468

Notes: The samples include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2016. To be included in the sample, principals must remain in their school at least 3 years. Thus, principals from entry cohorts in 2017 and 2018 will not contribute to these estimates.

Estimates are based on a two-way fixed effect regression with school fixed effect, year fixed effects, and grade indicators. Period t=0 which is the first year of the new principal's tenure is dropped from the sample.

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Appendix Table A4

Gardner (2022) two-stage event-study estimates (controlling for period before)

		Levels		Gains
		Math	Reading	Math	Reading
Pre-period	Resident^*(t=−5)	−0.0052 (0.0154)	−0.0121 (0.0105)	0.0090 (0.0086)	0.0005 (0.0057)
Pre-period	Resident^*(t=−4)	0.0098 (0.0135)	0.0094 (0.0087)	0.0179^* (0.0105)	0.0159^* (0.0073)
Resident^*(t=−3)	−0.0017 (0.0097)	0.0034 (0.0097)	−0.0075 (0.0104)	−0.0074 (0.0073)
Resident^*(t=−2)	0.0045 (0.0147)	0.0043 (0.0097)	−0.0095 (0.0101)	−0.0014 (0.0078)
Post-period	Reference period (t=−1)
	Resident^*(t=0) (dropped)
	Resident^*(t=1)	0.0435 (0.0321)	0.0333 (0.0252)	0.0433^*** (0.0157)	0.0258^** (0.0126)
Resident^*(t=2)	0.0687 (0.0436)	0.0438 (0.0287)	0.0418^** (0.0197)	0.0097 (0.0114)
Sample Size	440968	440968	352705	351864

Notes: Similar to Table 4, the sample here include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2018. Schools have at least 5 years of data prior to the entry and at least 2 years following the entry of the new principal. Note that schools where a new principal entered in 2018 will only have 2 “post-period” years.

Estimates are based on the two-stage approach in Gardner (2022), where the first stage uses untreated observations to estimate school and period effects, and these effects are then removed prior to comparing outcomes for treated and untreated observations. Regressions include school fixed effects and year fixed effects. Regressions also control for indicators for grade and an indicator for the period before the transition (t=−1). Period t=0 which is the first year of the new principal's tenure is dropped from the sample,

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.001.

Appendix Table A5

TWFE control coefficients

	Mathematics		Reading
	Levels	Gains	Levels	Gains
Student Black	−0.5523^***	−0.1834^***	−0.3968^***	−0.1482^***
0.0346	0.0112	0.0368	0.0111
Student Hispanic	0.3420^***	−0.1112^***	−0.327^***	−0.1047^***
0.0261	0.0083	0.0297	0.0094
Student Other Race	0.0989^*	0.0532^**	−0.0816^***	−0.0061
0.0569	0.0210	0.0279	0.0113
Student Female	−0.0434^***	−0.0061^***	0.1095^***	0.0312^***
0.0051	0.0019	0.0046	0.0019
Student Special Education	−0.0975	−0.2951^***	−1.2028^***	−0.3671^***
0.0175	0.0088	0.0145	0.0076
Sample Size	440968	352705	440968	351864

Notes: Similar to Table 4, the sample here include students in schools that hire a new principal (resident trained or non-resident trained) in the years 2014 to 2018. Schools have at least 5 years of data prior to the entry and at least 2 years following the entry of the new principal. Note that schools where a new principal entered in 2018 will only have 2 “post-period” years.

The control coefficient estimates come from regressions that include school fixed effects and year fixed effects. Regressions also control for indicators for grade. Period t=0 which is the first year of the new principal's tenure is dropped from the sample,

Standard errors clustered by school are in parentheses.

* p < 0.10,

** p < 0.05,

*** p < 0.001.

School district investments in general skills: The case of principal residency programs

Figures & Tables

Table 1

Table 2

Figure 1

Table 3

Table 4

Table 5

Table 6

Table 7

Appendix Table A1

Appendix Table A2

Appendix Table A3

Appendix Table A4

Appendix Table A5

Paradigm

My account