III.1. Empirical Models

First-stage regressions: First, we measure the impact of ME exposure on a mother’s education. This is the focus of our earlier work (Dang et al., 2021). We use two complementary measures of exposure to the ME program. The first measure of exposure is if a mother lived in Northern provinces under complete DRV-control during the First Indochina War (1946–1954) in her prime schooling age. The econometric specification takes the form:

where S_M is a mother’s education; L_M is a dummy variable that equals 1 if a mother was born between 1940 and 1945 (the treatment group) and equals 0 if she was born either before the war in 1924–1935 (the first control group) or after the war in 1950–1961 (the second control group)¹⁰. The mother in the treatment group would have been of primary school age during the war. We remove from the estimation sample the cohorts born during 1936–1939 or 1946–1949, who were partially exposed to increased school access (i.e., the “contaminated” treatment cohorts). ME_M is a dummy variable that takes the value of 1 if the mother was born in the three provinces in the North (Nghe An, Thanh Hoa, Ha Tinh) that were under complete DRV’s control during the War, and 0 if the mother was born in other Northern provinces. The control variables X_M include dummy variables indicating the mother’s ethnicity and religion (i.e., whether the mother is Buddhist, Christian, or has other religious beliefs). Mother’s birth province fixed-effects (bp_M) and mother cohort fixed-effects (τ_M) are included.

The use of the first exposure measure implies one assumption: there was no ME program in mixed-occupied provinces. A few mountainous Northern provinces fell to the DRV’s control in the middle of the War (see Appendix Figure A1). They are closer to communist China, a key supporter of the DRV government. The first exposure puts mixed-occupied provinces in the “control” provinces. Unfortunately, we do not have exact information regarding the ME programs in these provinces. To the extent that the ME program might have been executed, either fully or partially, in these mixed-occupied provinces, the first exposure measure would underestimate the impact of the ME program.

The second exposure measure is the number of years a mother was exposed to the ME program. The specification is as follows:

where Expose_M is the number of years a mother was exposed to the ME program when she was between 6 and 11 years old, the crucial years of her primary education. This exposure varies between 0 and 5. In other words, we explicitly account for the differential length of exposure to the ME program that school-age girls residing in DRV-controlled areas might have benefited from increased school access. This exposure varies depending on their birth years, and the year a province fell to DRV control during the war. Appendix Table A2 illustrates this exposure, where a mother born in 1939 or later in DRV-controlled areas would have had full five years of exposure¹¹. The number of years of exposure is similarly adjusted for children born in French-controlled provinces or provinces that DRV gained control from around the middle of the war. We restrict the number of years of exposure to five because children would more likely drop out of school if they miss the crucial years of schooling at the primary school level.¹² Our second exposure thus includes all the “contaminated” treatment cohorts in the estimation sample. The estimation sample includes all mother’s birth cohorts from 1924 to 1961. Other control variables are similarly defined as in equation (1).

Note that the second exposure carries an implicit assumption: in the provinces that fell to the DRV’s control in the middle of the War, ME programs restarted in the first year of the DRV occupation. However, given wartime, the ME programs might have been delayed or never carried out systematically in those provinces. If so, the second measure would imply systematic measurement errors and underestimate the ME program’s effect, as the first instrument does (see the discussion on the previous page)¹³.

We exclude from our estimations mothers who had migrated when they were younger than 15 years old in both regressions. This exclusion is to consider only mothers who stayed in the provinces they were born at least until they were 15 years old. Robust standard errors are adjusted for two-way clustering (mother’s birth province and household’s current commune).

Reduced-form regressions: Next, we examine if a mother’s exposure to the ME program would affect her children’s education. We follow the same setup above, using the two exposure measures:

Equations (3) and (4) show the reduced-form regressions, where the dependent variable ${\overline{Y}}_c$ represents her children’s average education. This variable is the main educational outcome. Two other outcomes are the education of her most educated child and the percentage of her children completing high school. Robust standard errors are also adjusted for two-way clustering (mother’s birth province and household’s current commune). X_M, bp_M, τ_M are similarly defined as above.

Instrumental Variable (IV) regressions: Finally, we produce IV estimates that capture the effect of a mother’s education on her children’s education. Maternal education is instrumented by exposure to the ME program.

where the dependent variable ${\overline{Y}}_c$ represents her children’s average education, the education of her most educated child, and the percentage of her children completing high school. $\delta \widehat{S_M}$ represents years of maternal education explained by exposure to the ME program, as shown in equations (1) and (2). Robust standard errors are adjusted for two-way clustering (mother’s birth province and household’s current commune). For the IV instrument to be valid, the exclusion restriction needs to hold, which states that the only way the ME affects child education is through their mother’s education. The main challenge regarding the identification strategy is to what extent the impact on the second generation’s education is via mother’s education. In general, the effects on the second generation’s education could work via a mother’s health, her economic and political connections, or other programs that the DRV also implemented during the war. In section IV.3, we address the challenge in detail.

III.2 Data

The first type of data is occupation during the First Indochina War. Since each province in Vietnam produced a book on the history of its local communist party and its army, our previous work, Dang et al. (2021), manually compiles data from these books to construct the years the DRV’s started to occupy a province. From the occupation data, we build the exposure measures to the ME program, namely whether a school-age child was exposed to it and the duration.

Possible issues might arise with the occupation data. During the war, the territorial divide was not likely as clear-cut. The occupation zones might not have been as exact and limited by the current provincial line. “Gray” zones might have existed, and local people could have been under partial control from both sides. In French control areas, guerilla warfare might have taken place.

Nevertheless, what is needed for our identification strategy is that the ME campaign was relatively more prevalent in treatment provinces than in control provinces.

The primary data set is the 1997–1998 round of the Vietnam Living Standards Survey (VLSS), conducted by the Vietnam General Statistic. The 1997–98 VLSS is nationally representative and collects data on about 6,000 households across the country. The survey covers rich data such as household members’ demographics, education, health, labor markets, household consumption, income, and assets. It also collects information on children of household members living regardless if they lived inside and outside the household at the survey time. A unique feature of this round of VLSS is to provide information on respondents’ date of birth, place of birth, whether they have moved from their birthplace or not, and the age at migration¹⁴. Another advantage of the dataset is that adult children in 1998 can be linked to their parents, even when they do not live with their parents. Appendix Table A1 provides summary statistics for the individuals in our sample.

The key variable, namely “years of education,” is calculated based on the highest grade that an individual completed. For instance, if an individual completed the fifth grade, the years of education would be 5. If an individual attended junior college, university, or post-graduate school, the years of education would be the sum of the highest school grade (which is 12) and the number of years the individual reportedly spent at junior college, university, or post-graduate school.

IV.1 ME Exposure and Mother’s Education (First-Stage)

We begin by graphically illustrating how the education of a mother born in DRV-controlled provinces differs from that of a mother born in French-controlled provinces before and after the ME program. Figure 1 presents years of maternal education with different control and treatment cohorts. The control cohorts include mothers born in 1924–1935 and 1950–1961. The treatment cohorts contain mothers who were born between 1940 and 1945. Years of maternal education in DRV-controlled areas and French-controlled areas move parallel during the first control period 1924–1935. However, the education gap between DRV-controlled and French-controlled provinces was much larger in the treatment cohort of 1940–1945. Years of maternal education in the DRV-controlled and the French-controlled provinces moved parallel again in the second control period 1950–1961. The visual suggests that mothers exposed to the ME program achieved more educational attainment than those who were not exposed to the ME program.

Years of Education for Girls and Boys Born in DRV-occupied Provinces versus Those Born in French-occupied Provinces

Citation: IMF Working Papers 2022, 245; 10.5089/9798400226922.001.A001

Notes: treated period is 1940–45, the control group period is 1924–1935 and 1950–1961, and the contaminated period is 1936–39 and 1946–49.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Years of Education for Girls and Boys Born in DRV-occupied Provinces versus Those Born in French-occupied Provinces

Citation: IMF Working Papers 2022, 245; 10.5089/9798400226922.001.A001

Notes: treated period is 1940–45, the control group period is 1924–1935 and 1950–1961, and the contaminated period is 1936–39 and 1946–49.

• Download Figure
• Download figure as PowerPoint slide

Years of Education for Girls and Boys Born in DRV-occupied Provinces versus Those Born in French-occupied Provinces

Citation: IMF Working Papers 2022, 245; 10.5089/9798400226922.001.A001

Notes: treated period is 1940–45, the control group period is 1924–1935 and 1950–1961, and the contaminated period is 1936–39 and 1946–49.

• Download Figure
• Download figure as PowerPoint slide

For completeness, we also show the lines for all male birth cohorts. The difference between those born in DRV-occupied provinces and those born in French-occupied provinces appears negligible.

Regression results not shown here corroborate the visuals: educational outcomes of school-age boys in DRV-occupied provinces and those in French-occupied provinces are small and not statistically significantly different. Note that there were already stronger priorities for boy education before the war.

The male enrollment rate was far higher than the female enrollment rate in the prewar period, making it harder for the ME program to improve the former than the latter. Available data from Nguyen (2013) and General Statistics Office (2004) show that the average prewar (1932–44) female enrollment rate was 4%, which is five times less than the corresponding male enrollment rate of 22%.

We now turn to the formal analysis of the impact of the ME program on female education, which is the focus of our previous work, Dang et al. (2021). The first-stage regressions are presented in Table 1 for the first (Panel A) and second (Panel B) exposure measures. The first measure of exposure is if a mother lived in provinces under complete DRV-control during the First Indochina War (1946–1954) in her prime school age. The second measure is the number of years a mother is exposed to the ME program. Section III.1 provides more detail.

^{Table 1.}

ME Exposure and Mother’s Education

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, birth year fixed effects and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 1.}

ME Exposure and Mother’s Education

	Reading Literacy	Math competency	Primary education completion or above	Secondary education completion or above	Years of education
	(1)	(2)	(3)	(4)	(5)
Panel A: First Exposure Measure
Treated	0.204*** (0.073)	0.192** (0.089)	0.263*** (0.065)	0.187** (0.084)	1.474*** (0.297)
N	1622	1622	1618	1618	1619
adj. R2	0.460	0.460	0.521	0.395	0.535
Panel B: Second Exposure Measure
Years of exposure to ME	0.028 (0.017)	0.021 (0.019)	0.043** (0.016)	0.023 (0.014)	0.253** (0.094)
N	1979	1979	1975	1975	1976
adj. R2	0.414	0.425	0.481	0.371	0.499

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, birth year fixed effects and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 1.}

ME Exposure and Mother’s Education

	Reading Literacy	Math competency	Primary education completion or above	Secondary education completion or above	Years of education
	(1)	(2)	(3)	(4)	(5)
Panel A: First Exposure Measure
Treated	0.204*** (0.073)	0.192** (0.089)	0.263*** (0.065)	0.187** (0.084)	1.474*** (0.297)
N	1622	1622	1618	1618	1619
adj. R2	0.460	0.460	0.521	0.395	0.535
Panel B: Second Exposure Measure
Years of exposure to ME	0.028 (0.017)	0.021 (0.019)	0.043** (0.016)	0.023 (0.014)	0.253** (0.094)
N	1979	1979	1975	1975	1976
adj. R2	0.414	0.425	0.481	0.371	0.499

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, birth year fixed effects and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

Outcome variables include reading literacy, math competency, primary education completion, secondary education completion, and years of education. The treatment effects are positive and statistically significant for different indicators of female educational attainment. Specifically, ME-mothers were between 19 to 26 percentage points higher in the probability of achieving reading, math literacy, and completing primary education or secondary education (Columns 1 to 4, Panel A) compared to their peers in French-controlled areas. ME-mothers also accomplished 1.5 more years of schooling than those residing in French-controlled areas (Column 5, Panel A), a sizable increase of 48 percent from the average 3.1 years of education in the pre-treatment period. Columns 1 to 4 of Panel B indicate that an additional year of exposure to the ME program leads to a 4 to 9 percent higher probability of reading and math competency and completing primary or secondary education. Column 5 of Panel B shows an additional year of exposure to ME primary education raises education by 0.25 years. Note that all estimation models exclude all girls who had migrated when they were younger than 15. This exclusion ensures we know the girls studied in their birthplaces, as we do not have information about the provinces they first migrated to.

As section III.1 discusses, both of the exposure measures could underestimate the effects of the ME program, because of the assumptions made about the ME-program at the mixed-occupied provinces in the mountainous North. As a robustness check, we remove the mixed-occupied provinces and report the results in Panel A3.1 of Appendix Table A3. The sample size drops but only slightly because the mixed-occupied provinces are in the mountainous North and have fewer households in the VLSS survey. Panel A3.1 shows that the findings remain robust with very quantitatively similar estimates. This robustness check provides some confidence about the two measures.

Is the differential education attainment due to the ME program? The short answer is yes. There could indeed be several potential contamination channels, such as differences in social-economic conditions of the DRV and the French-control areas and selection bias (such as systematic migration out of the country of children under French-control areas). Dang et al. (2021) address them in detail and provide support for the educational impact of the ME program. Most importantly, the educational outcomes of school-aged girls in DRV-controlled areas where the ME program was not implemented (in South Vietnam) were not better than those in French-controlled areas. The finding supports the argument that the ME program—rather than other DRV-related factors (e.g., idealism or political/economic connections)- resulted in better long-term educational outcomes for those who grew up in DRV-controlled regions. Another concern is that more educated school-age children residing in French-controlled areas may have more systematically migrated out of the country after the war. To check this hypothesis, Dang et al. (2021) provide estimates on whether the control group is more likely to receive remittances from their relatives living overseas at the time of the survey in 1998. Estimation results suggest that they do not.

IV.2 ME Exposure and the Next Generation’s Education

This section presents the first key finding of our paper. It provides the effects of a mother’s exposure to the ME program on children’s education. In other words, this is the reduced-form effect. We provide the estimation using three educational outcomes: children’s average years of education, years of education of the most educated child, and the percentage of children completing high school. We do not examine the percentage of children completing primary and middle school because primary and middle school education was almost universal in Vietnam during the 1990s. Note that all children living within and outside the household are included.

Panel A of Table 2 reveals that children of a ME-mother had an average of 0.9 more years of education. They are also 15 percent more likely to finish high school. Note that a ME-mother had 1.5 more years of education on average herself. Panel B shows that an additional year of a mother’s exposure to the ME program is associated with 0.19 more years of her children’s education and 3 percent more likely to finish high school. Note that the coefficients of Panels A and B are quite consistent with each other. The coefficients in Panel A are roughly 5 times the coefficients in Panel B. Recall that that Panel B uses the second exposure measure, where we set the maximum number of exposure years to 5 (see section III.1).

^{Table 2.}

Mother’s ME Exposure and Children’s Education

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 2.}

Mother’s ME Exposure and Children’s Education

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A: First Exposure Measure
Treated	0.894*** (0.249)	0.808*** (0.191)	0.149*** (0.039)
N	1206	1206	1206
adj. R2	0.347	0.385	0.125
Panel B: Second Exposure Measure
Years of exposure to ME	0.188** (0.076)	0.129 (0.076)	0.031*** (0.008)
N	1546	1546	1546
adj. R2	0.322	0.349	0.123

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 2.}

Mother’s ME Exposure and Children’s Education

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A: First Exposure Measure
Treated	0.894*** (0.249)	0.808*** (0.191)	0.149*** (0.039)
N	1206	1206	1206
adj. R2	0.347	0.385	0.125
Panel B: Second Exposure Measure
Years of exposure to ME	0.188** (0.076)	0.129 (0.076)	0.031*** (0.008)
N	1546	1546	1546
adj. R2	0.322	0.349	0.123

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

As a robustness check, we remove the mixed-occupied provinces and report the results in Panel A3.2 of Appendix Table A3. The findings remain robust with very quantitatively similar estimates. The only difference is now the coefficient capturing the impact of one more year of mother’s ME exposure on the education of the most educated child becomes statistically significant.

IV.3 On the Causal Impact of Maternal Education on Her Children’s Education

a. Discussions about Potential Confounding Factors

Exposure to the ME program helps improve mothers’ education. It also helps improve children’s education. Is the relationship between a mother’s education and children’s education causal? In this section, we argue that the relationship is causal. We go through our argument in steps.

Confounding factors associated with living under DRV-controlled areas (and not related to mothers’ education) could help the next generation obtain high education. For example, the DRV might have implemented other policies (besides the ME program) that affected the lives of young Vietnamese children and hence the way they raised their children. In addition, it is possible that living under the DRV during the First Indochina War could later give mothers better political and economic connections to boost their children’s education.

However, we find evidence that helps argue against this type of confounding factor. The evidence is that children of exposed fathers (i.e., ME-fathers) do not have significantly higher education than children of non-ME fathers. If the DRV implemented other policies besides the ME program relevant to the next generation’s education, we would expect exposed ME-fathers to affect children’s education positively. Similarly, if political or economic connections were to play a significant role in their children’s education, we would expect ME-fathers to have at least as great, if not greater, political and economic connections as ME-mothers to boost their children’s education. After all, politics and business in Vietnam have been dominated by males. As Table 3 indicates, we do not see such effects. A father’s exposure to the DRV regime and the ME program at the prime schooling age does not significantly affect his children’s education. Although the effects of ME-fathers on their children’s education are still positive, they generally have a smaller magnitude than those of ME-mothers, and they are not statistically significant. This finding, together with the finding that ME-fathers do not have significantly higher education than non-ME-fathers, is consistent with the notion that the effect on the next generation’s education operates via parental education. Note that we do not rule out the possibility of other policies by the DRV or political and economic connections. We only claim that we do not find evidence for the systematic impacts of such policies or connections on the next generation’s education.

^{Table 3.}

Father’s ME Exposure and Children’s Education

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, paternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Children are restricted to being at least 23 years old. Robust standard errors in parentheses are adjusted for two-way clustering (father’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 3.}

Father’s ME Exposure and Children’s Education

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A: First Exposure Measure
Treated	0.898 (0.531)	0.565 (0.502)	0.109 (0.097)
N	974	974	974
adj. R2	0.431	0.470	0.187
Panel B: Second Exposure Measure
Years of exposure to ME	0.121 (0.102)	0.069 (0.114)	0.015 (0.016)
N	1256	1256	1256
adj. R2	0.394	0.432	0.150

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, paternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Children are restricted to being at least 23 years old. Robust standard errors in parentheses are adjusted for two-way clustering (father’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 3.}

Father’s ME Exposure and Children’s Education

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A: First Exposure Measure
Treated	0.898 (0.531)	0.565 (0.502)	0.109 (0.097)
N	974	974	974
adj. R2	0.431	0.470	0.187
Panel B: Second Exposure Measure
Years of exposure to ME	0.121 (0.102)	0.069 (0.114)	0.015 (0.016)
N	1256	1256	1256
adj. R2	0.394	0.432	0.150

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, paternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Children are restricted to being at least 23 years old. Robust standard errors in parentheses are adjusted for two-way clustering (father’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

The argument presented above against these confounding factors would be invalid if fathers play little role in their children’s education and mothers play a much bigger role. However, this is not likely. The literature has pointed out the important roles of fathers in children’s education (see, for example, Behrman and Rosenzweig, 2002 and Suhonen and Karhunen, 2019).

Another concern is regarding gender-specific confounding factors. They include policies implemented by the DRV that only affected how school-age girls would educate their children but did not affect how school-age boys would. It is difficult to think of an example. In the unlikely possibility of such a policy, we find evidence that helps argue against this type of confounding factors. The evidence is as follows. In the South of Vietnam, three provinces were under complete DRV’s control during the war. However, the ME program was not systematically carried out there for various reasons¹⁵. As a result, southern DRV-exposed mothers do not have better educational outcomes than other non-exposed mothers (see Dang et al., 2021). Here, we find that children of southern DRV-exposed mothers do not have significantly higher education than children of non-exposed mothers (see Table 4). If other non-ME, gender-specific policies were carried out in these southern provinces, the next generation’s education should benefit as well.

^{Table 4.}

Mother’s DRV Exposure and Children’s Education in South Vietnam

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 4.}

Mother’s DRV Exposure and Children’s Education in South Vietnam

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A: First Exposure Measure
Treated	0.553 (1.018)	0.453 (0.925)	0.033 (0.063)
N	1313	1313	1314
adj. R2	0.268	0.295	0.060
Panel B: Second Exposure Measure
Years of exposure to ME	0.113 (0.229)	0.126 (0.203)	0.007 (0.012)
N	1658	1658	1659
adj. R2	0.261	0.289	0.048

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 4.}

Mother’s DRV Exposure and Children’s Education in South Vietnam

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A: First Exposure Measure
Treated	0.553 (1.018)	0.453 (0.925)	0.033 (0.063)
N	1313	1313	1314
adj. R2	0.268	0.295	0.060
Panel B: Second Exposure Measure
Years of exposure to ME	0.113 (0.229)	0.126 (0.203)	0.007 (0.012)
N	1658	1658	1659
adj. R2	0.261	0.289	0.048

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

Hence, access to schooling at the right age boosted ME-mothers’ education compared to non-exposed mothers who had to enroll in the ME program at older ages after the war. In turn, ME-mothers’ higher education attainment boosted their children’s education. Of course, we cannot completely rule out all other channels, but the falsification tests discussed above help argue against them.

b. Inter-generational Impact of Education

If ME mother’s education is the dominant channel of impact on children’s education, we can run IV estimations to estimate the intergenerational education impact. The estimated coefficient can be interpreted as “the upper bound” of the intergenerational impact.

The impact is shown in Table 5, with OLS and the two instruments corresponding to ME exposure measures. The OLS results show that years of maternal education are significantly correlated with all measures of children’s education. Higher years of maternal education are correlated with higher average and maximum years of education of her children and a larger percentage of her children completing high school. The correlation between maternal education and her children’s average education of 0.31 to 0.35 is not very far from the global average correlation between parent and child’s schooling of 0.4, estimated by Hertz et al. (2007).

^{Table 5.}

Impact of Maternal Education on Children’s Education

Note: Each cell presents the results from a separate regression that controls dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 5.}

Impact of Maternal Education on Children’s Education

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A. First instrument
OLS regressions
Years of maternal education	0.308*** (0.042)	0.245*** (0.043)	0.036*** (0.005)
IV regressions
Years of maternal education	0.583** (0.280)	0.528*** (0.179)	0.097** (0.043)
F-test of the excluded instrument	13.04	13.04	13.04
N	1206	1206	1206
Panel B. Second instrument
OLS regressions
Years of maternal education	0.350*** (0.033)	0.298*** (0.033)	0.041*** (0.005)
IV regressions
Years of maternal education	0.651** (0.312)	0.445** (0.213)	0.108** (0.050)
F-test of the excluded instrument	7.64	7.64	7.64
N	1545	1545	1545

Note: Each cell presents the results from a separate regression that controls dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 5.}

Impact of Maternal Education on Children’s Education

	Average years of children education	Years of education of the most educated child	Percentage of children completing high school
	(1)	(2)	(3)
Panel A. First instrument
OLS regressions
Years of maternal education	0.308*** (0.042)	0.245*** (0.043)	0.036*** (0.005)
IV regressions
Years of maternal education	0.583** (0.280)	0.528*** (0.179)	0.097** (0.043)
F-test of the excluded instrument	13.04	13.04	13.04
N	1206	1206	1206
Panel B. Second instrument
OLS regressions
Years of maternal education	0.350*** (0.033)	0.298*** (0.033)	0.041*** (0.005)
IV regressions
Years of maternal education	0.651** (0.312)	0.445** (0.213)	0.108** (0.050)
F-test of the excluded instrument	7.64	7.64	7.64
N	1545	1545	1545

Note: Each cell presents the results from a separate regression that controls dummy variables indicating gender, religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

All IV estimates have the same sign as the OLS estimates. However, they are highly significant and have a larger magnitude. The IV estimates suggest that one more year of schooling for the mother can raise her children’s average education by about 0.58 additional years if the first instrument is used, or about 0.65 additional years if the second instrument is used (Column 1). The magnitude of the coefficient of the impact of years of maternal education on years of children’s education is greater in IV estimates than in OLS estimates. The estimate represents both direct effects from mothers and indirect spillover effects from fathers, which could be driven by assortative mating. The estimate is larger than the previous magnitude reported (for example, Black et al. (2005) report 0.2 to 0.25 for Norway; Suhonen and Karhunen (2019) report 0.54 for Finland; and Agüero and Ramachandran (2020) report less than 0.1 for Zimbabwe). We will argue that this is due to the more limited labor market outcomes for educated individuals in Vietnam, allowing educated parents to spend more time at home and benefit their children’s education more. We will explain our argument in more detail in the next section.

V.1 Maternal labor market and reproduction outcomes

Family income is positively associated with children’s improved education (Masters, 1969; Mayer, 1997; Akee et al., 2010). Mothers with better education might have better income. As more financial resources are available, this may help improve their children’s education. Unfortunately, information on individual income is not available in rural Vietnam. Many household members were self-employed. Therefore, we cannot disaggregate household income by individuals. Nevertheless, diversifying into non-farm activities is an effective tool to move out of poverty and increase income in rural Vietnam (Hoang et al., 2014). We use different indicators of participation in non-farm activities to proxy for mother’s income: a dummy variable for maternal non-farm participation in the past seven days, a dummy variable for a mother having paid work in the past seven days, a dummy variable for a mother working as a white-collar worker, and a dummy variable for a mother having health insurance, an indication of having had formal work. Findings presented in the first four columns in Table 6 show that a woman’s exposure to the ME program does not affect her labor market outcomes.

^{Table 6.}

Mother’s ME Exposure, Maternal Labor and Reproduction Outcomes

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 6.}

Mother’s ME Exposure, Maternal Labor and Reproduction Outcomes

	Non-farm participation in the past seven days	Paid work	White-collar	Health insurance	First birth age	Number of children	Hours of household chores
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
First Exposure
Treated	0.035 (0.062)	-0.008 (0.024)	0.064 (0.074)	0.069 (0.096)	0.470 (0.604)	-0.025 (0.214)	1.054 (1.446)
N	1623	1623	1623	1623	1522	1622	1522
adj. R2	0.115	0.066	0.063	0.095	0.058	0.203	0.089
Second Exposure
Years of ME exposure	0.009 (0.014)	0.003 (0.006)	0.012 (0.015)	0.010 (0.014)	0.053 (0.146)	0.006 (0.053)	0.235 (0.271)
N	1980	1980	1980	1980	1869	1979	1857
adj. R2	0.119	0.069	0.064	0.102	0.058	0.206	0.093

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 6.}

Mother’s ME Exposure, Maternal Labor and Reproduction Outcomes

	Non-farm participation in the past seven days	Paid work	White-collar	Health insurance	First birth age	Number of children	Hours of household chores
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
First Exposure
Treated	0.035 (0.062)	-0.008 (0.024)	0.064 (0.074)	0.069 (0.096)	0.470 (0.604)	-0.025 (0.214)	1.054 (1.446)
N	1623	1623	1623	1623	1522	1622	1522
adj. R2	0.115	0.066	0.063	0.095	0.058	0.203	0.089
Second Exposure
Years of ME exposure	0.009 (0.014)	0.003 (0.006)	0.012 (0.015)	0.010 (0.014)	0.053 (0.146)	0.006 (0.053)	0.235 (0.271)
N	1980	1980	1980	1980	1869	1979	1857
adj. R2	0.119	0.069	0.064	0.102	0.058	0.206	0.093

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

A potential reason for the lack of evidence for improved labor market outcomes for more educated ME-mothers is that many of them may no longer be economically active in the 1997–1998 VLSS. The treatment cohort (i.e., born 1940–1945) had been in their late 50s. Nevertheless, even for the variable of having health insurance, an indication of having had formal work, ME-mothers were not significantly more likely to have health insurance more than non-ME-mothers.

A woman with a better education level would likely delay the first child (Black et al. 2008; Silles, 2011; Ozier, 2016; Grönqvist and Hall, 2013; DeCicca and Krashinsky, 2015). This delay may increase child education. Higher female education also reduces fertility (Osili and Long, 2008; Keats, 2018). Put differently, a woman with a better education would have fewer children. Having fewer children can lead to children’s higher education (Hanushek, 1992; Li et al., 2008). This literature suggests that maternal education may affect her children’s education through fertility mechanisms. To test this hypothesis, Columns 5 and 6 estimate the impact of ME exposure on age at first birth and the number of children. The reduced-formed regressions show statistically insignificant findings, suggesting the exposure to DRV occupation and the ME program did not change the mother’s reproduction outcomes.

Table 6 also reports the results on hours of a mother’s household chores such as cleaning the house, preparing meals, washing the family’s clothes, buying food or clothes, fetching water or firewood that a mother did in the past seven days (column [7]). Again, ME exposure measures do not change this outcome. This finding is consistent with the maternal labor market above, which shows that maternal education has no significant impact on labor market outcomes.

V.2 Paternal education, labor market outcome, and lifestyle

We turn next to investigating whether a mother’s ME exposure has any association with paternal education outcomes. A variety of paternal education measures are used: a dummy variable for reading literacy, a dummy variable for math competency, a dummy variable for primary education completion or above, a dummy variable for secondary education completion or above, and years of education. The reduced-form regressions reveal marginal effects on the husband’s education, especially reading literacy and math competency (see Table 7). Note that the education association between mothers and fathers could be driven by assortative mating and/or mutual encouragement and support after marriage.

^{Table 7.}

Mother’s ME Exposure and Father’s Education

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 7.}

Mother’s ME Exposure and Father’s Education

	Reading Literacy	Math competency	Primary education completion or above	Secondary education completion or above	Years of education
	(1)	(2)	(3)	(4)	(5)
First Exposure
Treated	0.075** (0.036)	0.090** (0.042)	0.052 (0.055)	-0.026 (0.059)	0.728 (0.474)
N	1155	1155	1157	1157	1157
adj. R2	0.176	0.144	0.226	0.241	0.245
Second Exposure
Years of ME exposure	0.013 (0.009)	0.016 (0.011)	0.005 (0.009)	-0.005 (0.014)	0.158 (0.114)
N	1425	1425	1427	1427	1427
adj. R2	0.162	0.148	0.214	0.230	0.246

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 7.}

Mother’s ME Exposure and Father’s Education

	Reading Literacy	Math competency	Primary education completion or above	Secondary education completion or above	Years of education
	(1)	(2)	(3)	(4)	(5)
First Exposure
Treated	0.075** (0.036)	0.090** (0.042)	0.052 (0.055)	-0.026 (0.059)	0.728 (0.474)
N	1155	1155	1157	1157	1157
adj. R2	0.176	0.144	0.226	0.241	0.245
Second Exposure
Years of ME exposure	0.013 (0.009)	0.016 (0.011)	0.005 (0.009)	-0.005 (0.014)	0.158 (0.114)
N	1425	1425	1427	1427	1427
adj. R2	0.162	0.148	0.214	0.230	0.246

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year, and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

We then turn to paternal labor market outcomes. The estimates of the impact of maternal education on labor market outcomes are reported in Table 8. A mother’s ME exposure does not significantly impact the father’s non-farm participation, paid work, the probability of working as a white-collar worker, and health insurance.

^{Table 8:}

Mother’s ME Exposure and Father’s Labor Market Outcomes

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 8:}

Mother’s ME Exposure and Father’s Labor Market Outcomes

	Non-farm participation in the past seven days	Paid work	White-collar	Health insurance
	(1)	(2)	(3)	(4)
First Exposure
Treated	0.041 (0.081)	0.007 (0.048)	0.097 (0.066)	0.106 (0.105)
N	1157	1157	1157	1157
adj. R2	0.159	0.080	0.083	0.147
Second Exposure
Years of ME exposure	0.014 (0.013)	0.004 (0.007)	0.016 (0.012)	0.010 (0.019)
N	1427	1427	1427	1427
adj. R2	0.159	0.081	0.080	0.151

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 8:}

Mother’s ME Exposure and Father’s Labor Market Outcomes

	Non-farm participation in the past seven days	Paid work	White-collar	Health insurance
	(1)	(2)	(3)	(4)
First Exposure
Treated	0.041 (0.081)	0.007 (0.048)	0.097 (0.066)	0.106 (0.105)
N	1157	1157	1157	1157
adj. R2	0.159	0.080	0.083	0.147
Second Exposure
Years of ME exposure	0.014 (0.013)	0.004 (0.007)	0.016 (0.012)	0.010 (0.019)
N	1427	1427	1427	1427
adj. R2	0.159	0.081	0.080	0.151

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

Similar to labor market outcomes of ME-mothers, there is little evidence for improved labor market outcomes of their husbands. The little evidence might be explained by the centrally planned nature of labor markets of Vietnam before the beginning of economic reforms in the late 1980s (Irvin, 1995).

This fact is in line with the ruling Communist Party’s efforts to build a socialist economy in North Vietnam during 1954–1975 and in the entire country after the country’s unification in 1975 (Cima, 1987; Dang, 2008). Labor allocations, especially in the public sector, were largely decided by the government (Dang, 1999). The formal private sector was virtually non-existent before the early 1990s (Schaumburg-Muller, 2000).

However, we do see strong evidence that husbands of ME-mothers had a healthier lifestyle. An educated woman may influence the husband’s behavior in positive ways. We use a husband’s smoking as a proxy to examine the impact of a mother’s education on her husband’s lifestyle. A mother’s ME exposure is not significantly correlated with her husband’s past smoking (column [1] of Table 9). However, it has significant and negative impacts on the likelihood that her husband is smoking now and the number of cigarettes the husband is smoking per day (columns [2] and [3] of Table 9), suggesting a good influence of educated wives. Fathers’ good lifestyle and involvement positively influence children’s social, behavioral, psychological, and educational outcomes (Sarkadi et al., 2008; Flouri and Buchanan, 2004).

^{Table 9.}

Mother’s ME Exposure and Father’s Lifestyle

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 9.}

Mother’s ME Exposure and Father’s Lifestyle

	Husband Ever-smoked	Husband Smoking now	Number of cigarettes a day by husband
	(1)	(2)	(3)
First Exposure
Treated	-0.005 (0.034)	-0.024** (0.010)	-0.142** (0.063)
N	1150	1150	1150
adj. R2	0.024	0.037	0.020
Second Exposure
Years of ME exposure	-0.001 (0.005)	-0.005*** (0.002)	-0.029* (0.015)
N	1420	1420	1420
adj. R2	0.035	0.040	0.031

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 9.}

Mother’s ME Exposure and Father’s Lifestyle

	Husband Ever-smoked	Husband Smoking now	Number of cigarettes a day by husband
	(1)	(2)	(3)
First Exposure
Treated	-0.005 (0.034)	-0.024** (0.010)	-0.142** (0.063)
N	1150	1150	1150
adj. R2	0.024	0.037	0.020
Second Exposure
Years of ME exposure	-0.001 (0.005)	-0.005*** (0.002)	-0.029* (0.015)
N	1420	1420	1420
adj. R2	0.035	0.040	0.031

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

V.3 Household’s Income, Expenditure, Assets, and Other Living Conditions

We next consider the impact of a mother’s ME exposure on households’ income and living standards. Income is typically not precisely measured in household surveys; hence, we also rely on other measures of living standards. Table 10 provides estimates for the impacts on household income, expenditure, assets (all are on a per capita basis and a logarithmic scale), and living conditions, namely living area per capita, house structure, and floor quality¹⁶.

^{Table 10.}

Mother’s ME Exposure and Household’s Living Conditions

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 10.}

Mother’s ME Exposure and Household’s Living Conditions

	Log of real income per capita	Log of real expenditure per capita	Log of real durable assets’ value per capita	Log of living area per capita	House structure	Floor quality
	(1)	(2)	(3)	(4)	(5)	(6)
First Exposure
Treated	0.298 (0.191)	0.153** (0.074)	0.269** (0.111)	0.046 (0.058)	0.056 (0.041)	0.174*** (0.032)
N	1494	1622	1611	1622	1623	1623
adj. R2	0.061	0.198	0.142	0.128	0.143	0.142
Second Exposure
Years of ME exposure	0.044 (0.030)	0.030** (0.013)	0.078*** (0.026)	0.009 (0.014)	0.024** (0.009)	0.046*** (0.009)
N	1838	1979	1966	1979	1980	1980
adj. R2	0.055	0.198	0.137	0.128	0.143	0.138

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 10.}

Mother’s ME Exposure and Household’s Living Conditions

	Log of real income per capita	Log of real expenditure per capita	Log of real durable assets’ value per capita	Log of living area per capita	House structure	Floor quality
	(1)	(2)	(3)	(4)	(5)	(6)
First Exposure
Treated	0.298 (0.191)	0.153** (0.074)	0.269** (0.111)	0.046 (0.058)	0.056 (0.041)	0.174*** (0.032)
N	1494	1622	1611	1622	1623	1623
adj. R2	0.061	0.198	0.142	0.128	0.143	0.142
Second Exposure
Years of ME exposure	0.044 (0.030)	0.030** (0.013)	0.078*** (0.026)	0.009 (0.014)	0.024** (0.009)	0.046*** (0.009)
N	1838	1979	1966	1979	1980	1980
adj. R2	0.055	0.198	0.137	0.128	0.143	0.138

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

The estimates reveal that ME-mothers’ households did not have significantly higher income per capita (column [1] of Table 10). This finding is consistent with prior findings that both ME-mothers and her husband did not have significantly better labor market outcomes at the survey in 1998. However, ME-mothers’ households had significantly higher expenditure per capita (column [2] of Table 10). Mother’s ME exposure boosted the household’s expenditure per capita by 15 percent and boosted the household’s assets per capita by 26 percent (column [3). Educated mothers’ households also lived in houses with better floor quality (columns [5] and [6]).

Why did households of more educated mothers not have significantly higher income but have significantly higher expenditure and assets? There can be several explanations. First, it could be that the income coefficient is simply not precisely estimated. Second, educated mothers’ past income might have been higher, although the current income is similar. Third, it is also possible that the higher living conditions are attributable to better political or economic connections that are not precisely captured by reported income (see, for example, the evidence presented by Markussen and Tarp, 2014). Finally, higher expenditure could be thanks to better lifestyle and expenditure decisions, which could lead to stronger asset accumulations over time, even with the same income. Table 9 already shows that husbands of more educated mothers were less likely to smoke now and had fewer cigarettes. Besides, in households of ME mothers, expenditure on cigarettes was sharply lower, and expenditure on education per kid aged 6 to 18 was sharply higher (Table 11). Note that these kids need not be the mothers’ children; they include all in households. The estimates imply ME exposure reduced household expenditure on cigarettes by 71 percent (column [1] of Table 11), while increased education expenditure per kid aged 6 to 18 years old by 37 percent.

^{Table 11.}

Mother’s ME Exposure and Expenses on Tobacco and Education

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

^{Table 11.}

Mother’s ME Exposure and Expenses on Tobacco and Education

	Log (expense on tobacco +1)	Educational expense per kid aged 6–18
	(1)	(2)
First Exposure
Treated	-0.710** (0.295)	0.369** (0.177)
N	1622	1144
adj. R2	0.038	0.134
Second Exposure
Years of ME exposure	-0.122* (0.068)	0.131** (0.049)
N	1979	1348
adj. R2	0.031	0.162

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.

View Table

^{Table 11.}

Mother’s ME Exposure and Expenses on Tobacco and Education

	Log (expense on tobacco +1)	Educational expense per kid aged 6–18
	(1)	(2)
First Exposure
Treated	-0.710** (0.295)	0.369** (0.177)
N	1622	1144
adj. R2	0.038	0.134
Second Exposure
Years of ME exposure	-0.122* (0.068)	0.131** (0.049)
N	1979	1348
adj. R2	0.031	0.162

Note: Each cell presents the results from a separate regression that controls for dummy variables indicating religion groups (including Buddhism and Christianity), whether the individual belongs to the major ethnic group, maternal birth year and birth province fixed effects. All estimation samples exclude those who had migrated when they were younger than 15. Robust standard errors in parentheses are adjusted for two-way clustering (mother’s birth province and household’s current commune); *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.