Heterogenous parental responses to education quality

ABSTRACT

Are parental inputs complements or substitutes to education quality? Using variation induced by identification into a gifted and talented (GT) program, I find no aggregate effects on parental behavior as a result of their child's access to a higher quality education. However, there are heterogeneous effects. Non-minority parents decrease engagement but increase tutoring. Minority and low-income parents increase engagement and increase both tutoring and in-home homework help. Results suggest that parental investments are not necessarily a strict complement or substitute but is nuanced dependent on demographic factors. I provide suggestive evidence that the primary mechanism is parental beliefs.

KEYWORDS:

• Parental investments
• parent engagement
• gifted
• tracking

JEL CODES:

• I21
• I28

Acknowledgments

The author would like to thank participants of the seminar at Teachers College, Columbia University, Babson College, and the annual conferences of the Association of Education Finance and Policy and Association of Public Policy Analysis and Management. He also thanks Peter Bergman, Judith Scott-Clayton, Scott Imberman, Alex Eble, Sarah Cohodes, Randall Reback, Thomas Bailey, Miguel Urquiola, Vivian Liu, and Rina Park for various levels of useful feedback and encouragement. All errors and opinions are my own.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Parent survey data used by school districts are generally done anonymously. In this case, the school district conducted the survey confidentially but ensured that a link can only be done through a multiple-step process. As a former survey administrator for a large district, I found this practice to be unique among districts.

2 Betts (2011) discusses much of the complexities in estimating the effects of tracking and ability-grouping programs.

3 Permission has not be granted to release the District name, so minor generalizations are made in informing the reader of the context, such as the rounding of demographic percentages.

4 Since the data obtained was across the course of over a decade, all demographics and processes are approximated and generalized to the most common numbers and practices during this timeframe. There are minor deviations – for example, there were a couple years where the percentage of low-income students were higher than normal.

5 In the United States, English learners or English language learners – often denoted as EL or ELL – are students who are not able to communicate sufficiently in English such that they can learn when materials are taught in English. These students generally require specialized education where language foundations are emphasized and subjects may be taught in multiple languages.

6 Appendix 9 shows some example questions from the assessment.

7 In an interview, the District program officer responded that they only remember using professional judgment ‘a handful of times’ over the past decade.

8 The process for obtaining the parent and student engagement survey results was cumbersome. Most survey data were in paper form and the data were entered manually. Two waves of data collection occurred for this study. The first was for my dissertation, which the District provided data already available electronically. Afterwards, an administrator noted that there were additional number, and years, of data available in hard copy. The second wave was primarily a manual data entry project.

9 In interviews with district officials, a recently retired district administrator said that there have been times when the district would slightly move the predetermined threshold upward as a result of last minute seat or budget limitations, and always to allow a few less students in than anticipated. However, there is no qualitative evidence that either students, teachers, or administrators were able to change the scores themselves. This does not invalidate the RDD, but it does mean that over time there might be evidence of bunching just under the threshold. For instance, if many students scored the exact same raw score, the district may choose to not offer seats to all of those students rather than to all those students. Over time, bunching may occur directly under the threshold, along with the perception of manipulation in density figures. However, given that there is no evidence that this is systematically done to include or exclude certain types of students, the assumption should still be valid. Additionally, as mentioned there is little difference in terms of baseline covariates across the threshold, which provides evidence for internal validity.

10 While administrative data includes language spoken at home, I cannot see whether families are immigrants of those particular countries.

11 Additional analysis and information on parental beliefs is given in the mechanism section.

12 There is little practical meaning to the control complier means for survey questions, since it is coded as a one if parents state ‘never’, a two if ‘rarely’, a three if ‘sometimes’, and a four if ‘often’. However, we can interpret a higher mean as an implication that parents spend more time helping child with homework.

13 For these estimates, the number of observations is much lower than the main tables. When the district allowed access to the data, they noted that for the years I was interested in the surveys were entirely on paper and needed to be inputted manually. Knowing that resources and time were limited and contributing to the fact that the survey were sorted by school, the research assistant and I randomized the schools we would collect data from. In all, we collected data for 33% of schools only for a few questions deemed necessary along with the identifiable information.

Additional information

Funding

The author is also grateful to Teachers College, Columbia University, and Babson College, which provided funding for data collection efforts. IRB approval was granted by Teachers College, Columbia University.