Abstract
The observed association between household income and children’s education cannot be interpreted as causal in the existence of potential income endogeneity. The non-natural factors often infringe on the orthogonality condition, even if they satisfy the relevance condition. Utilising a farm household survey, saline soil, and rainfall data from Thailand, this paper exploits natural experiments using actual natural factors as an instrumental variable to find valid instruments and check the robustness of the results using many estimators. The results show that, among the nine instrumental variables, rainfall amount and rainfall deviation are valid instruments for establishing the causal effect of income on a child’s education.
Acknowledgements
The author would like to thank Dr. Butsara Yongkhamcha, Department of Biology, Faculty of Science, Mahasarakham University, Thailand, for her helpful advice about saline soil knowledge. The author is grateful to the Office of Agricultural Economics (OAE) for providing the Thailand Socio-Economic Survey (SES) of agricultural household and labor data. The author also thanks the Meteorological Department of Thailand for providing rainfall data. In addition, the author thanks the Land Development Department (LDD) of Thailand which provides and explains saline soil data sources.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability statement
The data used in this article are available online: https://data.mendeley.com/drafts/3z5b52tpyg.
Notes
1 presents saline soil map in the northeast of Thailand.
2 This matching technique is the same as that used by Paxson (Citation1992), which matched households in the SES of NSO with rainfall data reported by 61 weather stations.
3 An example of the calculation of EC50 and EC150 is presented in .
4 The same sample households are used to examine the responses of households to natural events because it was necessary to control for the differences in household groups, which may be a factor in the variation of results. In other words, it was necessary to examine whether and how the same households respond to each instrument.
5 Shaw (Citation1999) sorts the EC1:5 into six levels: <0.07 (very low), 0.07–0.14 (low), 0.15–0.33 (medium), 0.34–0.62 (high), 0.63–0.92 (very high), and >0.93 (extreme).
6 The average annual rainfall from 1988 to 2006 in the south, central (including east and west), and the northeast and north was 2017, 1450, 1403, and 1277 mm, respectively.
7 Most researchers use two main measurements of rainfall as instruments. For example, Abiona and Koppensteiner (Citation2018), Jacoby and Skoufias (Citation1997), Kazianga and Udry (Citation2006), and Paxson (Citation1992), who used rainfall deviation as an instrument, while Bengtsson (Citation2010), Fichera and Savage (Citation2015), Lind (Citation2020), and Miguel and Satyanath (Citation2011) used either current or lag of rainfall as an instrument.
8 The advantages and disadvantages of these estimators are discussed in Andrews, Moreira, and Stock (Citation2007), Bascle (Citation2008), Hahn et al. (Citation2004), Ng and Bai (Citation2009), and Stock et al. (Citation2002).
9 Due to the space limitation, the first-stage regression was reported only in the case of using all IVs. The first-stage regression for other cases can be available upon request.
10 Both actual and subjective data on saline soil are still a weak instrument even though it may satisfy the exogeneity condition. Thus, this is not reported in the results for space reasons.
11 If the number of endogenous regressors is equal to the number of instruments, the bias is close to zero (Angrist & Krueger, Citation2001).
12 Because the model is a just-identified model, the validity of an instrument cannot be tested; thus, it is not reported in these cases.