In this section, we present the results from the specifications discussed in Section III. We begin by providing an overview of the average effect of school type on student performance across countries, followed by an examination of the specific results within each country.
7.1 Overall analysis
Table
2 displays the results for Eq. (
1) in columns (1), (2), and (3). For a detailed table containing all control variables, please refer to Table
A.2 in Online Appendix III. Across all three subjects, private schools show a significant positive correlation with students’ educational achievements. This translates into a 2.5% increase in mathematics and science scores and a 2.9% increase in reading scores, given that the mean score for the entire sample ranges from 470 to 476 points, depending on the subject.
Table 2
OLS estimates of the effect of the type of school on students’ scores. Math, reading and science
Private school | 11.63*** (4.355) | 13.51*** (4.480) | 11.77** (4.435) | | | |
Private gov-dependent | | | | 0.940 (3.284) | 3.411 (3.270) | 2.042 (3.377) |
Private independent | | | | 19.80*** (5.481) | 21.23*** (5.647) | 19.22*** (5.561) |
Control var. student | Yes | Yes | Yes | yes | Yes | Yes |
Control var. school | Yes | Yes | Yes | Yes | Yes | Yes |
Control school climate | Yes | Yes | Yes | Yes | Yes | Yes |
Country fixed effect | Yes | Yes | Yes | Yes | Yes | Yes |
Observations | 326,659 | 326,659 | 326,659 | 326,659 | 326,659 | 326,659 |
R-squared | 0.464 | 0.419 | 0.407 | 0.465 | 0.420 | 0.408 |
However, when controlling for students’ socio-economic background, the increase in scores associated with attending privately operated schools becomes more modest. Additionally, specification (1) explains between 41 and 46% of the total variation in the sample.
The complete table in Online Appendix III (Table
A.2) reveals interesting trends. On average, girls perform better than boys in reading, while boys tend to excel in science and mathematics. Starting school between the ages of 4 and 7 has a significantly positive connection with children’s performance compared to starting primary school at age 3. However, starting school after age 8 negatively affects student performance. Being schooled in a language different from one’s mother tongue reduces student scores, with this effect being nearly twice as large for reading and science compared to mathematics. This aligns with the fact that mathematics requires fewer language skills.
Parental education levels also have a positive effect on student achievement. Interestingly, the immigration status of the mother has a slightly significant positive connection with reading scores, while the father’s immigration status does not seem to affect outcomes. This discrepancy may be attributed to the diverse range of countries included in the analysis, where the impact of parental immigration status differs between more and less developed countries. Therefore, examining this coefficient across such a wide variety of contexts may not be meaningful. Further country-level analysis reveals that the coefficient for parental immigration status is consistently significant, though its direction varies across countries.
The number of books at home is another significant and strong predictor of student performance. Children with more than 100 books at home score approximately 50 points higher on average compared to those with fewer than 10 books. This represents an advantage of over 10%, supporting Wößmann’s (
2003) finding that the number of books at home is a reliable proxy for a child’s socio-economic and cultural background.
In terms of the cultural possession index, children from families that prioritize arts and culture tend to perform better on average. Additionally, students with access to a computer at home score significantly higher than those without. Among school characteristics, students perform better when the student-to-teacher ratio is lower, and those attending schools in larger towns or cities tend to achieve higher results. The size of the school has a small but positive effect on scores. Interestingly, class size also has a positive correlation with exam scores, as long as class sizes remain below 35 students.
School climate also plays a role. Disruptions caused by student truancy, skipping classes, or lack of respect negatively affect performance, while issues related to drugs and bullying appear to have no significant effect. Regarding teacher responsibility, if principals report that learning is disturbed by unprepared teachers, this leads to lower student outcomes. However, negative climate factors such as teachers being too strict or resistant to change do not seem to negatively affect performance; in some cases, the relationship is even positive.
When comparing publicly and privately financed public schools, columns (4) to (6) of Table
2 show that attending a government-dependent private school does not significantly raise student scores. However, attending private independent schools leads to approximately a 4% increase in results. Given Dronkers and Robert’s (
2008) findings on the positive impact of government-dependent private schools on student performance, we suspect that the lack of significance in our results may be due to controlling for school climate variables. To evaluate this, we reran the first and second specifications without controlling for school climate. These results, found in Table
A.3 in Online Appendix III, show that controlling for school climate partially mitigates the positive effect of private schools. The connection between private independent schools and reading and science scores remains significantly positive. These findings partly confirm Dronkers and Robert’s (
2008) conclusion that the advantage of private schools is partially due to better school climates. However, our results contradict the first hypothesis from past research, as we also expected a clearer influence from government-dependent private schools, which we did not observe.
In estimating regression (
3), we analyze how school composition affects student performance, as shown in Table
3. For the estimated coefficients of all control variables, please refer to Table
A.4 in Online Appendix III. Table
3 reveals that school composition significantly and substantially influences student performance. Increasing the mean ESCS (Index of Economic, Social, and Cultural Status) of a school by one point (equivalent to one standard deviation among OECD countries) results in an average increase of approximately 44 points in a student’s score—an increase of nearly 10% from the average score. These results confirm the second hypothesis presented in Sect.
4.
Table 3
OLS estimates of the effect of school composition on student’s scores. Math, reading and science
School composition | 43.14*** (2.606) | 44.72*** (2.738) | 43.39*** (2.728) |
Control var. student | yes | yes | yes |
Control var. school | yes | yes | yes |
Control school climate | yes | yes | yes |
Country fixed effect | yes | yes | yes |
Observations | 326,659 | 326,659 | 326,659 |
R-squared | 0.498 | 0.455 | 0.441 |
Regarding control variables, the coefficients remain relatively stable, with a slight decrease in the magnitude of the coefficients for family background variables that favor student performance. The coefficient for the student-to-teacher ratio is no longer significant. Specification (3) explains between 44 and 50% of the total variation in the sample.
Finally, as demonstrated in Figs.
1 and
2 of Sect.
5, private schools tend to enroll students from more favorable socio-economic backgrounds. Therefore, we include both school composition and school type in the regression, following specification (4). A summary of the results can be found in Table
4, and the estimated coefficients are in Table
A.5 in Online Appendix III. According to the estimations, the coefficients for private schooling are no longer significant. Conversely, the coefficient for school composition remains significant and substantial. This suggests that when the socio-economic composition of the school is considered, private schools do not outperform public schools.
Table 4
OLS estimates of the effect of the type and composition of the school on students’ scores. Math, reading and science
Private school | − 1.656 (4.312) | − 0.179 (4.338) | − 1.579 (4.402) | | | |
Private gov-dependent | | | | − 3.017 (3.172) | − 0.678 (3.233) | − 1.949 (3.288) |
Private independent | | | | − 0.512 (5.953) | 0.240 (6.158) | − 1.268 (6.141) |
School composition | 43.43*** (2.768) | 44.75*** (2.835) | 43.66*** (2.904) | 43.24*** (2.883) | 44.68*** (2.889) | 43.61*** (3.005) |
Control var. student | Yes | Yes | Yes | Yes | Yes | Yes |
Control var. school | Yes | Yes | Yes | Yes | Yes | Yes |
Control school climate | Yes | Yes | Yes | Yes | Yes | Yes |
Country fixed effect | Yes | Yes | Yes | Yes | Yes | Yes |
Observations | 326,659 | 326,659 | 326,659 | 326,659 | 326,659 | 326,659 |
R-squared | 0.498 | 0.455 | 0.441 | 0.498 | 0.455 | 0.441 |
In the fifth specification, which distinguishes between various sources of funding for private schools, we find that neither independent private schools nor government-dependent private schools show better performance compared to public schools with similar socio-economic indices. An interpretation of the results in Table
4 is that private and public schools themselves do not significantly shape exam performance. Instead, what really seems to matter is the socio-economic composition of the school.
Following Dronkers and Robert (
2008), we perform the same test as before, removing the controls for school climate. The estimated coefficients are shown in Table
A.6 in Online Appendix III. Removing these variables does not affect the significance of the coefficients for private schools. These results contradict the third hypothesis based on existing research. While Dronkers and Robert (
2008) emphasized the importance of controlling for school composition, they still found a positive impact of government-dependent private schools. However, in our analysis, when controlling for school composition, there is no longer any advantage to attending any type of private school.
This result is robust, as the coefficient for school composition remains unchanged across specifications (3), (4), and (5). Furthermore, comparing Table
4 and Table
A.6 in Online Appendix III, we observe that part of the link between school composition and exam results is that a higher average socio-economic level in a school contributes to a better school climate.
In conclusion, the above analysis suggests that the socio-economic composition of the school is a better predictor of performance and institutional quality than whether the school is public or private. However, these effects are likely to vary across countries. Therefore, we now estimate regressions for each country.
7.2 Analysis by country
Table
5 presents the estimations of country-specific regressions. A notable finding is that school composition is consistently significant and positive in all countries except one (Iceland). In contrast, the link between private schools and exam performance varies across countries, with some showing no effect and others displaying positive or negative correlations.
Table 5
OLS estimates of the effect of the type and composition of the schools on student’s scores by country. Scores in Mathematics
Algeria | 49.102*** | 50.976*** | 24.669*** | 0.243 | 3,455 | 0.008 | 0.003 |
Argentina | 32.646*** | 17.282 | 34.217*** | 0.540 | 1,094 | 0.286 | 0.193 |
Australia | − 5.497*** | 6.293** | 39.413*** | 0.294 | 10,290 | 0.274 | 0.129 |
Belgium (Fr) | 20.226*** | – | 24.559*** | 0.509 | 2,136 | 0.542 | 0 |
Brazil | 4.651 | 21.659*** | 28.068*** | 0.332 | 10,147 | 0.004 | 0.101 |
Bulgaria | – | − 12.886 | 74.213*** | 0.460 | 4,984 | 0 | 0.010 |
Canada | 32.279*** | 22.372*** | 19.522*** | 0.224 | 14,701 | 0.025 | 0.040 |
Chile | 3.233 | 7.341* | 33.844*** | 0.473 | 5,511 | 0.428 | 0.264 |
China (B-S-J-G) | 48.525*** | − 4.835 | 56.559*** | 0.393 | 9,002 | 0.008 | 0.090 |
China (Hong Kong) | − 9.102** | – | 40.619*** | 0.321 | 4,520 | 0.922 | 0 |
China (Macao) | − 9.910 | − 33.339** | 31.903*** | 0.264 | 4,221 | 0.834 | 0.139 |
Chinese Taipei | − 51.638*** | − 53.081*** | 74.133*** | 0.408 | 7,071 | 0.058 | 0.239 |
Colombia | − 14.421*** | 10.703*** | 33.651*** | 0.327 | 8,212 | 0.063 | 0.191 |
Costa Rica | 16.109*** | 2.474 | 24.814*** | 0.280 | 5,598 | 0.022 | 0.103 |
Croatia | 8.909 | 0.000 | 80.670*** | 0.389 | 5,200 | 0.023 | 0.001 |
Czech Republic | − 7.788** | − 46.758*** | 72.087*** | 0.524 | 5,921 | 0.071 | 0.009 |
Denmark | 4.394 | 17.846** | 11.519*** | 0.287 | 4,881 | 0.140 | 0.020 |
Dominican Republic | − 6.574 | − 7.029* | 35.040*** | 0.374 | 2,780 | 0.039 | 0.180 |
Estonia | − 14.953** | − 31.469*** | 46.904*** | 0.237 | 4,761 | 0.027 | 0.008 |
Finland | 6.809 | – | 34.912*** | 0.220 | 5,179 | 0.044 | 0 |
France | 2.323 | 6.439 | 53.707*** | 0.521 | 4,382 | 0.136 | 0.066 |
Georgia | 8.406 | 32.853*** | 30.076*** | 0.305 | 3,577 | 0.014 | 0.037 |
Germany | − 0.307 | 0.000 | 69.505*** | 0.493 | 3,496 | 0.066 | 0.006 |
Greece | – | 6.445 | 41.252*** | 0.301 | 4,550 | 0 | 0.054 |
Hungary | − 10.598*** | 4.875 | 66.020*** | 0.563 | 4,671 | 0.163 | 0.022 |
Iceland | 24.686 | – | 9.408 | 0.187 | 2,775 | 0.005 | 0 |
Indonesia | − 11.260*** | − 16.276*** | 47.087*** | 0.430 | 4,462 | 0.220 | 0.142 |
Ireland | 11.064*** | 5.152 | 19.563*** | 0.300 | 4,429 | 0.524 | 0.026 |
Italy | − 5.073 | − 21.907*** | 66.152*** | 0.343 | 7,345 | 0.064 | 0.022 |
Japan | − 53.274*** | − 38.317*** | 118.001*** | 0.459 | 6,091 | 0.035 | 0.272 |
Jordan | − 19.145** | − 2.799 | 23.719*** | 0.214 | 5,027 | 0.004 | 0.217 |
Korea | 1.800 | 12.686*** | 64.625*** | 0.334 | 5,303 | 0.221 | 0.107 |
Kosovo | 115.191*** | 36.383*** | 26.692*** | 0.262 | 3,584 | 0.007 | 0.035 |
Latvia | 19.001* | − 1.145 | 20.009*** | 0.225 | 4,227 | 0.007 | 0.008 |
Lebanon | − 45.937*** | 5.443 | 33.212*** | 0.339 | 1,981 | 0.009 | 0.481 |
Lithuania | 18.401** | 21.159* | 45.143*** | 0.280 | 5,856 | 0.013 | 0.004 |
Luxembourg | − 20.796*** | 20.556** | 48.420*** | 0.448 | 4,443 | 0.091 | 0.024 |
Macedonia | 0.000 | 16.418* | 44.535*** | 0.302 | 3,304 | 0.007 | 0.024 |
Malta | 16.438 | 30.434** | 38.764*** | 0.300 | 2,294 | 0.231 | 0.162 |
Mexico | – | − 9.541*** | 24.702*** | 0.238 | 6,250 | 0 | 0.098 |
Moldova | − 47.593** | 30.599*** | 23.207*** | 0.239 | 4,014 | 0.003 | 0.014 |
Montenegro | – | 0.000 | 92.228*** | 0.318 | 5,093 | 0 | 0.005 |
Netherlands | 6.494** | 0.000 | 74.207*** | 0.549 | 2,174 | 0.577 | 0.000 |
Norway | − 12.601 | – | 9.093* | 0.196 | 3,865 | 0.014 | 0 |
Peru | 26.899*** | 12.485*** | 26.152*** | 0.395 | 5,890 | 0.022 | 0.245 |
Poland | 17.110* | 25.854** | 28.282*** | 0.264 | 3,842 | 0.020 | 0.010 |
Portugal | − 9.493 | 6.631 | 21.775*** | 0.278 | 5,263 | 0.008 | 0.032 |
Qatar | 17.917** | 15.710*** | 76.234*** | 0.483 | 9,212 | 0.018 | 0.353 |
Romania | – | 1.969 | 60.270*** | 0.402 | 4,691 | 0 | 0.011 |
Russian Federation | 15.584 | − 25.053* | 33.243*** | 0.138 | 5,172 | 0.002 | 0.002 |
Singapore | 23.292** | − 71.605*** | 58.385*** | 0.377 | 4,865 | 0.018 | 0.036 |
Slovak Republic | − 2.962 | – | 51.506*** | 0.445 | 5,816 | 0.104 | 0 |
Slovenia | − 20.220*** | – | 92.962*** | 0.427 | 5,155 | 0.020 | 0 |
Spain | − 3.703 | − 6.356 | 16.127*** | 0.274 | 5,548 | 0.283 | 0.057 |
Switzerland | − 5.648 | − 47.700*** | 49.372*** | 0.418 | 4,636 | 0.021 | 0.027 |
Trinidad and Tobago | 10.490 | − 49.464*** | 80.074*** | 0.526 | 2,819 | 0.040 | 0.022 |
Tunisia | – | − 54.019*** | 36.628*** | 0.364 | 3,390 | 0 | 0.008 |
Turkey | − 23.244** | − 53.473*** | 50.019*** | 0.436 | 5,425 | 0.006 | 0.035 |
United Arab Emirates | − 28.817*** | 1.399 | 58.564*** | 0.399 | 9,002 | 0.014 | 0.518 |
United Kingdom | 13.794*** | 1.828 | 47.752*** | 0.306 | 7,912 | 0.017 | 0.043 |
United States | – | − 21.406*** | 26.169*** | 0.285 | 4,524 | 0 | 0.053 |
Uruguay | – | 2.936 | 36.155*** | 0.397 | 5,124 | 0 | 0.177 |
Vietnam | – | − 33.628*** | 37.847*** | 0.332 | 5,516 | 0 | 0.075 |
In this section, we provide general observations on the potential effects of private schools across countries and then focus on school composition, drawing comparisons with Figs.
1 and
2 from Sect.
5.
First, Table
5 reveals significant variation in schooling systems across countries, making it challenging to draw general conclusions about the impact of school type on student scores. Out of the 63 countries in the sample, only 16 show a positive and significant correlation between private government-dependent schools and student scores. These countries are geographically diverse and have varying percentages of private government-dependent schools, making it difficult to pinpoint the factors driving the premium associated with this type of school. However, once we control for school composition, the advantage of private schools primarily stems from the socio-economic characteristics of their student populations. In 37 out of 63 countries, private schools do not show a positive advantage over public schools in terms of student scores when controlling for school composition.
Next, we examine the role of school composition and its potential association with segregation induced by private schooling, comparing the findings with Figs.
1 and
2. Iceland is the only country where school composition does not seem to affect exam performance. In Norway, school composition has a relatively small and marginally significant effect. It is worth noting that these two countries have minimal differences in the average socio-economic status of private and public schools, as shown in Fig.
2. Denmark, on the other hand, shows weak but highly significant effects of socio-economic segregation on student outcomes, where a one-point increase in the socio-economic index corresponds to an average score increase of 11.52 points.
As Fig.
2 indicates, Asian countries (excluding Singapore) show no or limited socio-economic segregation between private and public schools. However, the positive and significant coefficient of school composition suggests that some form of socio-economic segregation still influences student scores, independent of school type. Italy and the Netherlands exhibit a similar pattern, with no significant difference in average school composition between private and public schools (Fig.
1). Nonetheless, the positive, significant, and substantial connection between school composition and student performance implies the presence of segregation unrelated to school type or differences in schooling systems.
Conversely, Latin American countries show wide disparities in average school composition between private and public schools (Fig.
2). The positive and significant coefficient of school composition indicates that the type of school in these countries likely generates segregation and inequalities in educational outcomes and future opportunities. This pattern also holds for Greece and Bulgaria, the two European countries with the largest socio-economic gap between the two school types.
In Belgium, private government-dependent schools exhibit a higher socio-economic composition, as seen in Fig.
1. Combined with the positive and significant coefficients for school composition and private schools, this suggests that students attending these schools enjoy significant advantages compared to those in public schools.