![Sample our Education journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5932/TOC-Subject-Sample-2021-Education.jpg"})
Abstract
Private tutoring has become popular throughout the world. However, evidence for the effect of private tutoring on students' academic outcome is inconclusive; therefore, this paper presents an alternative framework: a nonparametric bounds method. The present examination uses, for the first time, a large representative data-set in a European setting to identify the causal effect of self-initiated private tutoring. Under relatively weak assumptions, I find some evidence that private tutoring improves students' outcome in reading. However, the results indicate a heterogeneous and nonlinear effect of private tutoring, e.g. a threshold may exist after which private tutoring becomes ineffective or even detrimental.
Acknowledgements
I would like to thank the Consortium PISA.ch for allowing recording the data and the Leading House on the Economics of Education, Firm Behaviour and Training Policies for support. I am grateful to Charles F. Manski, David Figlio, Stefan Boes, and Stefan C. Wolter, participants of the IWAEE Conference, as well as the referee for thoughtful comments.
Notes
1 Only studies that control in a credible way for the endogeneity of private tutoring are included in this literature review.
2 The equivalent of states in the USA.
3 Because of item non-response, 2383 observations were deleted.
4 To make the notation more compact, I leave the conditioning on covariates (x) and the notation for mathematics (m) and reading (r) implicit in the following.
5 It is important to notice that these bounds are not confidence intervals. They express the ambiguity created by the selection problem (Manski and Pepper, Citation2011).
6 The ATE (E[y(1)] − E[y(0)]) is calculated as follows: The lower bound on E[y(1)] minus the upper bound on E[y(0)] is the lower bound of the average treatment effect. The upper bound on E[y(1)] minus the lower bound on E[y(0)] is the upper bound of the ATE.
7 Manski bounds are sharp bounds, i.e. nothing else can be learned in face of the censored data (see the proof in Heckman and Leamer, Citation2007; Heckman and Vytlacil, Citation2000; Manski, Citation2007).
8 Furthermore, it could be that ability and taste for private tutoring are positively associated. Therefore, more able students would want to go to further lessons after school.
9 High-scoring students are students with a PISA competence level 6, and low-performing students are those with a PISA competence level equal to or below 1.
10 For example, Ono (Citation2007) uses tutoring during secondary education as in IV to measure the effect of tutoring in tertiary education.
11 The identifying power of an MIV is examined in Manski and Pepper (Citation2000).
12 The MIV used is discrete and takes four possible values: no post-obligatory education, vocational education, secondary academic education, and tertiary education.
13 For proof, see Manski and Pepper (Citation2000).
14 Self-learning time was not questioned in PISA 2009. For this present research, the questions in the international PISA student questionnaire concerning tutoring are not detailed enough to distinguish between private (and privately paid) tutoring and other out-of-school time lessons. Comparing participation rates in tutoring (international question) and private tutoring (national option on privately paid tutoring) shows an overestimation in the international question. The international question leads to participation rates of 40% (OECD, Citation2011b) in tutoring compared with 30% participation rate in privately paid tutoring.
15 The data at hand do not allow an analysis on the number of weekly hours spent on private tutoring.
16 Manski and Pepper (Citation2011) applied the method of restricting the minima and maxima.
17 Because of the small numbers for some MIV values, calculation of MIV-bounds for high and low performing students is not possible.
