Heterogeneity: Hometown Ethnicity Distribution

In “belief-based” models, people who possess less prior belief would be more af-fected by the new information. (DellaVigna and Gentzkow,2010). We test this prediction by splitting into two groups: people who lived in towns with more Hoklo people, and otherwise.¹⁴ This division distinguishes the familiarity of stu-dents to Taiwan-related knowledge. Children may randomly pick up

cultural parents/role models in his living neighborhood (Bisin et al.,

2011). Since Hoklo people in the older generation typically hold stronger Taiwanese identity, the ra-tio of Hoklo people surrounding the respondents regulates the probability for a junior high school student picking a role model with strong Taiwanese identity.

Furthermore, in terms of political discussions such as election campaigns, people living in towns with less Hoklo people would be exposed to less Taiwan-oriented speeches since politicians running for local elections have to cater to local peo-ple’s political preference including national identity.

To show that our criterion distinguishes local environments with different na-tional identity trend. We utilize 1992, 1995 1998 and 2000 waves of TSCS and calculate the mean of

Identity in High Hoklo Ratio Towns and Low Hoklo Ratio Towns (divided by the Hoklo people population median in 2004) in different

sur-vey years. The results are graphed in Figure10. It is obvious that the difference between the proportion of people holding stronger Taiwanese identity in these two areas is of size about 0.1 to 0.15. This assures us that students lived in two areas face greatly different society atmosphere in terms of issues regrading

na-14We use the variableHome-Hoklo-Ratio, which assigns 1 if the proportion of Hoklo people in one’s hometown is less than the population median (77.1%), zero otherwise, to divide the sample.

We provide the map demonstrating the Hoklo ethnicity distribution based onHome-Hoklo-Ratio in AppendixE.

tional identity when in junior high school and elementary school, the time when they absorb the input for national identity formation from the living environ-ment. We argue that the textbook effect would be greater for students living in

Low Hoklo Ratio Towns according to “belief-based” models, since they were

orig-inally less familiar with Taiwan-related knowledge.

We first look at the raw national identity variation in different survey years.

In Fig 11, we observe that in 2005 wave, respondents who lived in

Low Hoklo Ratio towns are much more likely to hold stronger Taiwanese identity than in

2004 wave, while other respondents who lived in

High Hoklo Ratio towns do not

exhibit such pattern. Despite this change from 2004 to 2005 waves, the jumps of the

Low Hoklo Ratio line around the cut-off are salient in both survey years and

the magnitude of the jump is bigger than the

High Hoklo Ratio line, which fits our

prediction from the belief-based model. We caution that the sample size of the cells of 1998 education cohort in 2004 is small (4 and 11 for

High Hoklo Ratio and Low Hoklo Ratio towns respectively), so one should not make interpretation on

the dip of the two dots in 2004. Using regression with specification2to test if the differences between the magnitudes of the jumps in two subgroups is significant, we do not, however, find significant results in Figure12.

To examine the finer variation within an education cohort, in figure 13 we plot the mean of

Identity at birth quarter level (controlling for survey fixed effect)

for observations lived in

Low Hoklo Ratio Towns (Figure

13a) and

High Hoklo Ratio

Towns (Figure

13b). A clear jump with the rough size 0.35 around the birth month cut-off is exhibited in Figure 13a, while in Figure 13b, the dots does not show meaningful variation.

In Panel A of Table 5, we find that the estimated textbook effect obtained from regression results including only respondents who lived in

Low Hoklo Ratio Towns with specification

1 is robust across different columns. The size is about 0.35 and significant, consistent with Figure13and larger than estimates in Table 2 (overall sample). For respondents who lived in

High Hoklo Ratio Towns, no

significant textbook effect is detected in Panel B of Table5. Although we observe significant heterogeneity in this dimension, we note that the dummy variable

Home-Hoklo-Ratio suffers from discontinuity around the birth month cut-off (see

Table1), which might cause potential problem in RD estimations in Table5.

The Relationship between Main Results and Subgroup Analysis.

Results of textbook effect along two dimensions, education track and Hoklo people proportion, display significant heterogeneity, The textbook effect interacts with the students´familiarity with the education content, both before (Home−

Hoklo

− Ratio heterogeneity) and after (Education Track heterogeneity).

The finding of textbook effect heterogeneity further helps us rule out two con-cerns. First, the measurement concern. One might suspect that the main results come from that students in earlier cohorts are afraid of responding their true na-tional identity. If this is the case, then the subgroup analysis would suggest that only students in academic education track and students who lived in

Low Hoklo

Ratio Towns were afraid of responding themselves as Taiwanese, which is very

unlikely. For the same reason, if the main results come from other factors such as significant social events, the event should affect students in all subgroups. We are thus more confident that the effect we detect in main results comes from the introduction of the new textbook–Knowing Taiwan series.

.2.3.4.5.6(mean) identity_use

1992 1995 1998 2000

Survey Year

High Hoklo Ratio Area Low Hoklo Ratio Area

Figure 10: National Identity Trend in High/Low Hoklo Ratio Areas in 1990s

Notes: We pool data from 1992, 1995, 1998 and 2000 TSCS waves. We drop the observations aged below 30 (to include those adults whom children are more likely to meet). We collapse the data into survey-year-Home-Hoklo-Ratio level. Dots represent the mean of Identity in each cell.

.3.4.5.6.7.8Taiwanese Identity

1995 1996 1997 1998

Education Cohort

Low High

2004

.3.4.5.6.7.8Taiwanese Identity

1995 1996 1997 1998

Education Cohort

Low High

2005

Figure 11: National Identity Trend in High/Low Hoklo Ratio Areas

Notes: Data is collapsed at education-cohort-survey year level for observations in High/Low Hoklo Ratio Areas from 2004 and 2005 waves of TSCS. Dots represent the simple mean of Identity at each cell.

-0.40-0.200.000.200.40Taiwanese Identity

1995 1996 1997 1998

Education Cohort

Point estimate 95% CI

Figure 12: Heterogeneity Textbook Effect in High/Low Hoklo Ratio Areas: Co-hort Analysis

Notes: We pool 2003, 2004 and 2005 TSCS waves. Point estimates and confidence intervals of β_2k are derived from regression with specification2reckoning 1996 education cohort as reference group. Standard errors are clustered at birth month level.

(a) Low Hoklo Ratio Area

-.4-.20.2.4

-10 -5 0 5 10

Birth Quarter

(b) High Hoklo Ratio Area

-.2-.10.1.2

-10 -5 0 5 10

Birth Quarter

Figure 13: Heterogeneity Textbook Effect in High/Low Hoklo Ratio Areas: RD visulization

Notes: We pool data from 2003, 2004, 2005 TSCS and include education cohorts from 1995 to 1998. Figure13aincludes observations responded lived longest inLow Hoklo Ratio Towns (Home-Hoklo-Ratio equal to 1) before age 15. Figure13bincludes observations responded lived inHigh Hoklo Ratio Towns (Home-Hoklo-Ratio equal to 0)longest before age 15. We first regress Identity on survey year dummies and then collapse the residuals at birth quarter level to derive the dots.

Fitted lines are from regression of the dots on a first order polynomial of birth quarter interacted withKnowingT aiwan dummy variable.

Table 5: Heterogeneity Textbook Effect: Hometown Ethnicity Dis-tribution

Dep. Variable:

Identity

(1) (2) (3) (4)

Panel A: Hometown with Low Hoklo People Ratio

KnowingT aiwan

0.336*** 0.366*** 0.392** 0.423***

(0.125) (0.125) (0.164) (0.155) Control Group Mean 0.555 0.555 0.555 0.555

Persuasion Rate 75.5 82.2 88 95.7

Observations 186 186 186 186

R-squared 0.147 0.222 0.159 0.230

Panel B: Hometown with High Hoklo People Ratio

KnowingT aiwan

0.0311 0.0321 0.0970 0.0790 (0.0976) (0.0980) (0.133) (0.148) Control Group Mean 0.647 0.647 0.647 0.647

Persuasion Rate 8.81 9.09 27.4 22.3

Observations 231 231 231 231

R-squared 0.137 0.213 0.141 0.215

Hometown FE Yes Yes Yes Yes

Survey FE Yes Yes Yes Yes

Linear Trend Yes Yes No No

Quadratic Trend No No Yes Yes

Demographic No Yes No Yes

Notes: We pool 2003, 2004, 2005 TSCS data. Panel A includes obser-vations whose hometown has lower Hoklo people ratio comparing to the median of the population in National Hakka Population Basic In-formation Survey Research, while Panel B includes observations whose hometown has higher ratio. Specifications in each column are the same as in corresponding columns in Table2. Standard errors clustered at birth month level in parentheses. *** p<0.01, ** p<0.05, * p<0.1

6 Long Run Results

We have until this section established that the introduction of new textbooks,

Knowing Taiwan series, has impacts on students’ national identity formation when

they were 18 to 23 years old (short run sample). The natural question emerges:

whether the impact is transitory or persistent.Clots-Figueras and Masella(2013) measure the Catalan identity and observe the voting behavior when samples are roughly 25 to 30 years old.Fouka(2015) used ethnic intermarriage and off-spring first names to measure the strength of one’s ethnic identity. Both paper found sig-nificant language policy effect in such long run outcomes. We explore this issue by examining the long run sample (surveyed from 2010 to 2015, respondents age 24 to 33).

We first look at the education cohort variation in Figure 14. Surprisingly, al-though the 1998 education cohort exhibits stronger Taiwanese identity, the 1997 education cohort is not significantly more likely to identify himself as Taiwanese.

We again examine the finer variation within education cohorts. In Figure15we plot the

Identity mean at birth quarter level (controlling survey year fixed effects).

No significant jump is found, and the mean level of

Identity is about the same on

either side of the cut-off. The regression results in Table6fit the graph and shows no sign of textbook effect in the long run. We note that the size of the coefficients of

TextbookExposure are all small across specifications, compared to the results

derived from short run sample.

This result, whether the students were exposed to the new textbooks cannot predict the long run national identity difference, prompts us to ask that which group’s national identity change in the long run, the control, the treatment, or

both. We aim to see that whether it is the students who were not exposed to the textbook are more likely to hold Taiwanese identity, or it is the students who were exposed to the textbook are less likely to hold Taiwanese identity in the long run. Figure 16 offers an interesting answer: national identity of students not exposed to the new textbook

catch up with the students who were exposed

to the new textbook.¹⁵ In Figure 16, we plot the simple mean of the

Identity

dummy variable in treatment group (people born in 1997 and 1998 education cohort) and control group (people born in 1995 and 1996 education cohort). We find that the proportion of people holding stronger Taiwanese identity in control group increases obviously from 2009 to 2015, when the respondents are around their thirties.

One hypothesis of why the national identity in both groups converge in the long run is the peer interaction. Although students who studied the old textbook were originally endowed with weaker Taiwanese identity comparing to students exposed to the new textbook (in the short run), the “old textbook” effect may be lessened by the peer interaction because the former student encounter more and more "Taiwanese-oriented" peers surrounding their living environment. Another hypothesis is that the control group is affected by the overall national identity trend illustrated in Figure 1: an increasing trend of Taiwanese identity. To ex-plore if any one of these two are more aligned with the data is a potential research direction.

15We do not use the data from 2003 in the following analysis since the sample size of treatment group in 2003 is too small. We include the 2009, 2015 data in the following analysis. For 2009 data, it is because we now do not require the parents’ education variables. For 2015 data, I will incorporate it in other parts of analysis in the future.

-0.20-0.100.000.100.20Taiwanese Identity

1993 1994 1995 1996 1997 1998

Education Cohort

Point estimate 95% CI

Figure 14: Estimated Textbook Effect by Education Cohort in the Long Run

Notes: We pool 2010, 2012, 2013, 2014 TSCS waves. Point estimates derived from regression ofIdentity on survey year dummies and education cohort dummies (1996 as reference group).

Standard errors are clustered at birth month level.

-.2-.10.1

-10 -5 0 5 10

Birth Quarter

Figure 15: Long Run National Identity Variation at Birth Quarter Level

Notes:I pool data from 2010, 2012, 2013, 2014 and 2015 TSCS waves and include education co-hort from 1995 to 1998. We first regressIdentity on survey year dummies and then collapse the residuals at birth quarter level to derive the dots. Fitted lines are from regression of the dots on a first order polynomial of birth quarter interacted withKnowingT aiwan dummy variable.

.5.6.7.8.9Taiwanese Identity

2005 2010 2015

Survey Year

Control Group Treatment Group

Figure 16: National Identity Variation by Treatment Status

Notes:I pool data from 2004, 2005, 2010, 2012, 2013, 2014 and 2015 TSCS and include education cohort from 1995 to 1998. Dots represent simple mean ofIdentity of each cell in different survey years. Control group includes 1995 and 1996 education cohorts and Treatment group includes 1997 and 1998 education cohorts

Table 6: Estimated effects of the textbook reform in the Long Run Dep. Variable:

Identity

[1] [2] [3] [4]

T extbookExposure

-0.0256 -0.0313 -0.0510 -0.0661 (0.0667) (0.0692) (0.0992) (0.107)

Month

-0.00240 -0.000606 0.0129 0.0153

(0.00205) (0.00251) (0.0133) (0.0149)

T extbookExposure

0.00958** 0.00614 -0.0172 -0.0194

×Month (0.00443) (0.00476) (0.0198) (0.0215)

Control Mean 0.78 0.78 0.78 0.78

Observations 535 535 535 535

R-squared 0.057 0.161 0.061 0.164

Hometown FE Yes Yes Yes Yes

Survey FE Yes Yes Yes Yes

Demographic No Yes No Yes

Notes: We pool the 2010, 2012, 2013, 2014 and 2015 TSCS waves and include two education cohorts on both sides of the birth-month cut-off (Sep 1984). Spec-ifications are the same as Table2. Clustered standard errors at birth month level in parentheses.*** p<0.01, ** p<0.05, * p<0.1

7 Conclusion

Can education content influence national identity formation? Although the effect of different education policy (e.g., language use and curriculum reform) on vari-ous preferences has been found in recent literature (Clots-Figueras and Masella, 2013;Fouka,2015;Cantoni et al.,2015). Direct relationship between the two has not been established yet. We consider the textbook reform which introduced new textbooks containing much more Taiwan-oriented materials (Knowing Taiwan se-ries). We adopt education cohort comparison and regression discontinuity design (RDD) on birth month level to tease out the national identity society trend. We find large and significant effect of the education content on holding stronger Tai-wanese identity.

We interpret the result coming from the priming effect of “Taiwan” and the instillation of Taiwan-related knowledge. Our results suggest that education con-tent could not only influence the preference over political institutions (Cantoni et al.,2015), but also national identity, providing more empirical evidence of the effect of education on homogenizing people, as advocated in theory literature (Alesina and Reich,2015). Nonetheless, since we measure the national identity by self response of survey questions instead of real behavior, the next step would be examining the effect of education content upon behavior such as voting.

We contribute to the literature by presenting three facts. First, the degree of exposure of education content matters. Students who studied the new text-book harder, measured by the academic/vocational track division, are more af-fected by the new textbook. Second, the education content effect is larger for students holding weaker prior beliefs (identified by the hometown ethnicity

dis-tribution), align with the belief based models in persuasion literature. Whether people holding stronger or weaker prior beliefs are more affected by education policies are not certain in the literature. Voigtländer and Voth (2015) finds that people who held stronger prior anti-Semitic attitude (identified by the support of anti-Semitic parties in previous election) are more affected by the anti-Semitic in-doctrination between 1933 and 1945; they exhibit largest increases in anti-Jewish attitudes. Why the persuasion is effective in different subgroups under different contexts is a potential research question in the future. The two findings, the im-portance of degree of exposure and the prior belief, generalize the findings in Cantoni et al.(2015), which documents the textbook effect in Peking university students, a group of students potentially more affected by the textbooks due to their hard work in senior high school.

Finally, we document that the textbook effect is not permanent by showing that whether the students study the new textbook could not predict significant national identity differences in the long run. We propose peer interaction channel and exposure to society trend as explanation but more examination is required.

Based on our finding, the long run effect of the curriculum reform in China stud-ied byCantoni et al.(2015), which changes students ideology significantly when they are still enrolled in Peking University, is thus an exciting topic for future research.

References

Aghion, P., Jaravel, X., Persson, T., and Rouzet, D. (2014). Education and military rivalry.

Working Paper.

Alesina, A. and Fuchs-Schundeln, N. (2007). Good-bye lenin (or not?): The effect of communism on people’s preference.

American Economic Review, 97(4):1507–

1528.

Alesina, A. and Reich, B. (2015). Nation building.

Working Paper.

Bisin, A., Patacchini, E., Verdier, T., and Zenou, Y. (2011). Formation and persis-tence of oppositional identities.

European Economic Review, 55(8):1046–1071.

Cantoni, D., Chen, Y., Yang, D. Y., Yuchtman, N., and Zhang, Y. J. (2015). Cur-riculum and ideology.

(forthcoming) Journal of Political Economy.

Clots-Figueras, I. and Masella, P. (2013). Education, language, and identity.

The Economic Journal, 123:F332–F357.

DellaVigna, S. and Gentzkow, M. (2010). Persuasion: Empirical evidence.

Annual Review of Economics, pages 643–669.

DellaVigna, S. and Kaplan, E. (2007). The fox news effect: Media bias and voting.

The Quarterly Journal of Economics, 122(3):1187–1234.

Fan, E., Liu, J.-T., and Chen, Y.-C. (2014). Is the ’quarter of birth’ endogeneous?

evidence from one million sibilings in taiwan.

NBER Working Paper Series.

Fouka, V. (2015). Backlash: The unintended effects of language prohibition in us schools after world war i.

Working Paper.

Liu, M.-H., Huang, L.-C., and Vickers, E. (2005).

History Education and National Identity in East Asia, chapter Identity Issues in Taiwan’s History Curriculum.

Routledge Taylor Francis Group, 1 edition.

Shigeoka, H. (2014). The effect of patient cost sharing on utilization, health, and risk protection.

American Economic Review, 104(7):2152–2184.

Voigtländer, N. and Voth, H.-J. (2015). Nazi indoctrination and anti-semitic be-liefs in germany.

Proceedings of the National Academy of Sciences of the United States of America, 112(26):7931–7936.

Wang, F.-C. (2001). National imagination, ethnic consciousness, and history:

Content and context analysis of the “getting to know taiwan” textbook dis-putes,.

Taiwan Historical Research, 8(2):145–208.

Appendices

A National Identity Distribution and Sample Selec-tion

Table 7: National Identity Distribution in Main Regression Sample Identity Taiwanese Both (Taiwanese and Chinese) Chinese Total

#Obs 269 132 16 417

Percent 64.51 3.84 31.65 100

Notes:

We pool 2003 2004 2005 TSCS waves and include two education cohorts on both sides of the birth-month cut-off (Sep 1984). The identity is measured by the re-sponse of two questions:

1. "In our society, somebody call themselves "Taiwanese," somebody call them-selves "Chinese," and somebody call themthem-selves "both." Do you consider yourself as "Taiwanese," "Chinese," or "Both"?".

2. "Below are some possible descriptions of yourself. Which is most suitable to you?

We define Taiwanese as respondents answering Taiwanese, Chinese to those an-swering Chinese, and Both to those anan-swering Both, Taiwanese and Chinese, Chi-nese and TaiwaChi-nese.

Table 8: Samples Dropped by Sample Selection

Short Run Sample Long Run Sample

Rule #Obs #Obs

Reports born outside Taiwan or 0 10

lived longest outside Taiwan before 15

Lack of Demographic Info 11 45

Answer Other in identity measure question 8 14

Total drop ratio 4.5% 12.8%

Final sample Size 417 535

Notes: The short run sample pools data from 2003, 2004, 2005 TSCS waves and the long run sample pools 2010, 2012, 2013, 2014 waves. We include two education cohorts on each side of birth month cut-off.

B Observable Characteristics

T able 9: Observ ables C on tin uity with P ol ynomials of Second Order V ARIABLES g ender F ather Mother F ather Mother F ather Mother Self Self Hometown Ethnicity Ethnicity Ed u Lev el Ed u Lev el Ed u Y ears Ed u Y ears Ed u T rack Ed u Y ears Hokl o Ra tio

TextbookExposure

-0.0188 -0.0975 -0.0172 -0.366*** -0.125 -1.589 -0.648 0.0358 -0.578* 0.256** (0.0847) (0.101) (0.0766) (0.134) (0.123) (1.050) (0.676) (0.0862) (0.310) (0.105)

Month

0.0170 0.0108 -0.00198 0.0387** 0.0217 0.238 0.156* 0.00890 0.0521 -0.0247 (0.0132) (0.0135) (0.0142) (0.0174) (0.0130) (0.152) (0.0894) (0.0119) (0.0493) (0.0164)

Month2

0.000643 0.000339 -8.37e-05 0.00149** 0.000884* 0.00841 0.00531 0.000228 0.00316 -0.000815 (0.000512) (0.000645) (0.000645) (0.000577) (0.000455) (0.00509) (0.00327) (0.000501) (0.00205) (0.000566)

TextbookExposure

-0.0266 0.00653 0.0187 0.0169 -0.0185 -0.00608 -0.135 -0.0468** 0.0216 0.0100 ×

Month

(0.0219) (0.0219) (0.0197) (0.0237) (0.0240) (0.182) (0.151) (0.0219) (0.0929) (0.0221)

TextbookExposure

-0.000451 -0.00144 -0.000966 -0.00386*** -0.000947 -0.0198*** -0.00626 0.00144 -0.00630 0.00121 ×

Month2

(0.00107) (0.000945) (0.000865) (0.000900) (0.00101) (0.00662) (0.00647) (0.00102) (0.00439) (0.000883) C onstan t 0.528*** 0.297*** 0.165*** 0.703*** 0.548*** 11.67*** 10.53*** 0.589*** 14.24*** 0.291*** (0.0697) (0.0523) (0.0588) (0.114) (0.0766) (0.999) (0.481) (0.0504) (0.225) (0.0971) Observ ations 417 417 414 417 417 417 417 417 417 417 R -squared 0.005 0.009 0.008 0.031 0.005 0.018 0.007 0.011 0.017 0.011

Notes

: W e pool da ta from 2003, 2004, and 2005 TSC S w av es. W e run regression with specifica tion

1

. Second order pol ynomial is incl uded and w e do not incl ude demogr aphics.

TextbookExposure

is 1 if the birth mon th of the responden t is after Sep 1984, 0 otherwise, indica ting if one studies the

KnowingTaiwan

series.

Month

is the birth mon th recen tered a t Sep 1984. The construction of the v ariables: Gender: female being 1, male being 0. F ather/Mother ethnicity: Hokl o fa thers/mothers being 0, otherwise 1. F ather /Mother ed uca tion lev el: fa ther/mother with no ed uca tion, elemen tary school, and junior high school ed uca tion lev el being 0. F ather/Mother/Self ed uca tion y ears is discrete v ariable rang e from 0 to 16. Self ed uca tion tr ack: responden ts with final ed uca tion a ttainmen t as senior high school, univ ersity being 1, otherwise, 0. Hometown Hokl o Ra tio is the d umm y v ariable

Home-Hoklo-Ratio

, construction see section

3

The proxy of hometown ethnicity distribution . Cl ustered standard error a t birth mon th lev el in paren theses.*** p<0.01, ** p<0.05, * p<0.1

(a) Female Ratio

0.2.4.6.81Father Education More than Junior High Ratio

-20 -10 0 10 20

Birth Month

(e) Mother Education Level

0.2.4.6.81Mother Education More than Junior High Ratio

-20 -10 0 10 20

(a) Mother Education Years

Notes: We pool 2003 2004 2005 TSCS waves and include education cohorts from 1995 to 1998.

We first regress the corresponding demographic variable on survey year dummies and then col-lapse the residuals at birth month level to derive the dots. Fitted lines are from regression of the dots on a first order polynomial of birth month interacted withT extbookExposure dummy variable. The construction of the independent variables is the following: Gender: female being 1, male being 0. Father/Mother ethnicity: Hoklo fathers/mothers being 0, otherwise 1. Father /Mother education level: father/mother with no education, elementary school, and junior high school education level being 0. Father/Mother/Self education years is discrete variable range from 0 to 16. Self education track: respondents with final education attainment as senior high school, university being 1, otherwise, 0. Hometown Hoklo Ratio is the dummy variable Home-Hoklo-Ratio, see construction in section3, The proxy of hometown ethnicity distribution.

C Bandwidth Sensitivity

-0.100.000.100.200.300.40

24 26 28 30 32 34 36 38 40 42 44 46 48

bandwidth

Point estimate 95% CI

Figure 19: Bandwidth Sensitivity with Demographic Controls

Notes: We run regressions as column [2] in Table2with different bandwidths. The bandwidth 48 means we include 24 birth months on each side of the cut-off, i.e., two education cohorts. The solid line represent the point estimate of theT extbookExposure dummy variable and the dotted line represents the corresponding 95% confidence interval derived from standard errors clustered at birth-month level.

(a) Bandwidth: 30 Months

-.6-.4-.20.2.4

-15 -12 -9 -6 -3 0 3 6 9 12 15

Birth Month

(b) Bandwidth: 28 Months

-.6-.4-.20.2.4

-15 -12 -9 -6 -3 0 3 6 9 12 15

Birth Month

(c) Bandwidth: 26 Months

-.6-.4-.20.2.4

-15 -12 -9 -6 -3 0 3 6 9 12 15

Birth Month

Figure 20: National Identity Variation at Birth Month Level with Different Band-widths

Notes: We collapse the residuals at birth month level after regressing Identity on survey year fixed effects. Each marker represent the simple mean of the residuals in each birth month. Fitted lines are derived from regression of the dots using linear polynomials of birth month interacted with

Notes: We collapse the residuals at birth month level after regressing Identity on survey year fixed effects. Each marker represent the simple mean of the residuals in each birth month. Fitted lines are derived from regression of the dots using linear polynomials of birth month interacted with

在文檔中 課綱與國族認同： 探討1997 年國中教科書改革的影響 (頁 49-72)