------------------------------------------------------------------------------- name: log: /Users/carlosmendez/Documents/GitHub/starter-academic-v501/content > /post/stata_did/analysis.log log type: text opened on: 26 Apr 2026, 11:26:07 . . di _newline(2) . di "============================================" ============================================ . di " Difference-in-Differences (DiD) Tutorial" Difference-in-Differences (DiD) Tutorial . di " Corral & Yang (2024)" Corral & Yang (2024) . di " $S_DATE $S_TIME" 26 Apr 2026 11:26:07 . di "============================================" ============================================ . . . *=================================================== . * PART 1: THE 2x2 DiD DESIGN . * Dataset: tutoring_did.dta (35 schools x 2 periods) . *=================================================== . . . *--------------------------------------------------- . * Section 1: Load and explore the 2x2 DiD dataset . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 1: DATA LOADING & EXPLORATION" SECTION 1: DATA LOADING & EXPLORATION . di "========================================" ======================================== . . use "https://github.com/quarcs-lab/data-open/raw/master/isds/tutoring_did.dta > ", clear . . * Inspect variable labels and storage types . describe Contains data from https://github.com/quarcs-lab/data-open/raw/master/isds/tuto > ring_did.dta Observations: 70 Variables: 7 20 Jul 2025 14:00 ------------------------------------------------------------------------------- Variable Storage Display Value name type format label Variable label ------------------------------------------------------------------------------- id float %9.0g School ID time float %9.0g Year (Intial = 1, Final = 2) treated float %9.0g Group membership (Treated = 1, Nontreated = 0) post float %9.0g Post policy period (Yes = 1, No=0) txp float %9.0g Treated x Post gpa float %9.0g School GPA (average GPA of low-income students) female_share float %9.0g Share of female students in the school ------------------------------------------------------------------------------- Sorted by: . . * Check means, SD, min, max . summarize Variable | Obs Mean Std. dev. Min Max -------------+--------------------------------------------------------- id | 70 18 10.17243 1 35 time | 70 1.5 .5036102 1 2 treated | 70 .2857143 .4550158 0 1 post | 70 .5 .5036102 0 1 txp | 70 .1428571 .3524537 0 1 -------------+--------------------------------------------------------- gpa | 70 77.11754 10.87694 59.39085 99.15061 female_share | 70 .5279242 .0265513 .4712442 .5695165 . . * Show a few rows for context . list in 1/6 +--------------------------------------------------------+ | id time treated post txp gpa female~e | |--------------------------------------------------------| 1. | 1 1 0 0 0 72.38992 .501643 | 2. | 1 2 0 1 0 81.30275 .5432417 | 3. | 2 1 0 0 0 69.89978 .5266564 | 4. | 2 2 0 1 0 81.39222 .562699 | 5. | 3 1 0 0 0 72.65748 .5476224 | |--------------------------------------------------------| 6. | 3 2 0 1 0 82.16635 .5110617 | +--------------------------------------------------------+ . . * Declare panel structure: id = school, time = period . xtset id time Panel variable: id (strongly balanced) Time variable: time, 1 to 2 Delta: 1 unit . . * Panel summary: within/between variance, balancedness . xtsum Variable | Mean Std. dev. Min Max | Observations -----------------+--------------------------------------------+---------------- id overall | 18 10.17243 1 35 | N = 70 between | 10.24695 1 35 | n = 35 within | 0 18 18 | T = 2 | | time overall | 1.5 .5036102 1 2 | N = 70 between | 0 1.5 1.5 | n = 35 within | .5036102 1 2 | T = 2 | | treated overall | .2857143 .4550158 0 1 | N = 70 between | .4583492 0 1 | n = 35 within | 0 .2857143 .2857143 | T = 2 | | post overall | .5 .5036102 0 1 | N = 70 between | 0 .5 .5 | n = 35 within | .5036102 0 1 | T = 2 | | txp overall | .1428571 .3524537 0 1 | N = 70 between | .2291746 0 .5 | n = 35 within | .269191 -.3571429 .6428571 | T = 2 | | gpa overall | 77.11754 10.87694 59.39085 99.15061 | N = 70 between | 1.124764 74.90352 79.88043 | n = 35 within | 10.81948 57.84736 96.38773 | T = 2 | | female~e overall | .5279242 .0265513 .4712442 .5695165 | N = 70 between | .0156944 .4964049 .5614545 | n = 35 within | .0214995 .4787881 .5770604 | T = 2 . . di _newline . di "Panel: 35 schools x 2 time periods = 70 observations" Panel: 35 schools x 2 time periods = 70 observations . di "Treatment: 10 schools receive after-school tutoring" Treatment: 10 schools receive after-school tutoring . di "Comparison: 25 schools do not" Comparison: 25 schools do not . . . *--------------------------------------------------- . * Section 2: Treatment visualization (panelview) . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 2: TREATMENT VISUALIZATION" SECTION 2: TREATMENT VISUALIZATION . di "========================================" ======================================== . . panelview gpa txp, i(id) t(time) type(treat) /// > prepost bytiming /// > xtitle("Time Period") ytitle("School ID") /// > legend(position(6)) /// > name(panelview_2x2, replace) # Variable # Missing % Missing -------------------------------------------- 1 gpa 0 0.0 2 txp 0 0.0 Missing for | how many | variables? | Freq. Percent Cum. ------------+----------------------------------- 0 | 70 100.00 100.00 ------------+----------------------------------- Total | 70 100.00 Note: White cells represent missing values/observations in data. . . graph export "stata_did_panelview_2x2.png", replace width(2400) (file stata_did_panelview_2x2.png not found) file stata_did_panelview_2x2.png written in PNG format . . di "Figure saved: stata_did_panelview_2x2.png" Figure saved: stata_did_panelview_2x2.png . . . *--------------------------------------------------- . * Section 3: Interrupted Time Series (ITS) -- Figure 1 . * Naive pre/post comparison for treated group only . * Shows why a simple comparison overstates the effect . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 3: INTERRUPTED TIME SERIES" SECTION 3: INTERRUPTED TIME SERIES . di " (Figure 1 -- Treated Group Only)" (Figure 1 -- Treated Group Only) . di "========================================" ======================================== . . preserve . collapse (mean) gpa, by(time treated) . . twoway (connected gpa time if treated==1, /// > msymbol(O) mcolor(gs1) lcolor(gs1) /// > ylab(0(10)100) xlab(1(1)2)), /// > ytitle("GPA") xtitle("Time") /// > xline(1.5, lcolor(red) lpattern(dash)) /// > title("Figure 1: Interrupted Time Series (Treated Group Only)") /// > note("Source: Corral & Yang (2024). Simulated data.") /// > graphregion(color(white)) plotregion(color(white)) /// > name(fig1_its, replace) . . graph export "stata_did_its.png", replace width(2400) (file stata_did_its.png not found) file stata_did_its.png written in PNG format . restore . . di "Figure saved: stata_did_its.png" Figure saved: stata_did_its.png . di _newline . di "Naive ITS comparison:" Naive ITS comparison: . di " Treated group GPA jumped from ~60 to ~96" Treated group GPA jumped from ~60 to ~96 . di " Naive change: ~36 GPA points" Naive change: ~36 GPA points . di " BUT: this ignores secular time trends!" BUT: this ignores secular time trends! . di " We need a comparison group to isolate the causal effect." We need a comparison group to isolate the causal effect. . . . *--------------------------------------------------- . * Section 4: Parallel Trends & Counterfactual -- Figure 2 . * Shows treated, control, and counterfactual trends . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 4: PARALLEL TRENDS" SECTION 4: PARALLEL TRENDS . di " (Figure 2 -- Counterfactual)" (Figure 2 -- Counterfactual) . di "========================================" ======================================== . . preserve . collapse (mean) gpa, by(time treated) . . * Compute counterfactual: what would treated look like without treatment? . * Counterfactual = treated_pre + control_change . quietly sum gpa if treated==0 & time==1 . local ctrl_pre = r(mean) . quietly sum gpa if treated==0 & time==2 . local ctrl_post = r(mean) . local ctrl_change = `ctrl_post' - `ctrl_pre' . . quietly sum gpa if treated==1 & time==1 . local treat_pre = r(mean) . local cf_post = `treat_pre' + `ctrl_change' . . di "Control pre: `ctrl_pre'" Control pre: 71.21514129638672 . di "Control post: `ctrl_post'" Control post: 82.10103607177734 . di "Control change: `ctrl_change'" Control change: 10.88589477539063 . di "Treated pre: `treat_pre'" Treated pre: 60.16577911376953 . di "Counterfactual post: `cf_post'" Counterfactual post: 71.05167388916016 . . * Add counterfactual observations (treated==2 for dashed line) . * After collapse, dataset has: time, treated, gpa (4 rows) . local N = _N . insobs 2 (2 observations added) . replace time = 1 in `=`N'+1' (1 real change made) . replace time = 2 in `=`N'+2' (1 real change made) . replace treated = 2 in `=`N'+1' (1 real change made) . replace treated = 2 in `=`N'+2' (1 real change made) . replace gpa = `treat_pre' in `=`N'+1' (1 real change made) . replace gpa = `cf_post' in `=`N'+2' (1 real change made) . . twoway (connected gpa time if treated==1, /// > msymbol(O) mcolor(gs1) lcolor(gs1)) /// > (connected gpa time if treated==0, /// > msymbol(+) mcolor(gs5) lcolor(gs5)) /// > (connected gpa time if treated==2, /// > msymbol(O) mcolor(gs1) lcolor(gs1) lpattern(shortdash_dot)), /// > ylab(0(10)100) xlab(1(1)2) /// > legend(order(1 "Treated" 2 "Comparison" 3 "Counterfactual")) /// > ytitle("GPA") xtitle("Time") /// > xline(1.5, lcolor(red) lpattern(dash)) /// > title("Figure 2: DiD Design with Counterfactual Trend") /// > note("Source: Corral & Yang (2024). Dashed line = counterfactual (treated > without program).") /// > graphregion(color(white)) plotregion(color(white)) /// > name(fig2_counterfactual, replace) . . graph export "stata_did_counterfactual.png", replace width(2400) (file stata_did_counterfactual.png not found) file stata_did_counterfactual.png written in PNG format . restore . . di "Figure saved: stata_did_counterfactual.png" Figure saved: stata_did_counterfactual.png . . . *--------------------------------------------------- . * Section 5: DiD Means Table -- Table 1 . * Manual calculation of the 2x2 DiD estimate . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 5: DiD MEANS TABLE (Table 1)" SECTION 5: DiD MEANS TABLE (Table 1) . di "========================================" ======================================== . . * Means table by treatment status and time period . table treated post, stat(mean gpa) nformat(%12.2f) ---------------------------------------------------------------------------------------- | Post policy period (Yes = 1, No=0) | 0 1 Total -----------------------------------------------+---------------------------------------- Group membership (Treated = 1, Nontreated = 0) | 0 | 71.22 82.10 76.66 1 | 60.17 96.37 78.27 Total | 68.06 86.18 77.12 ---------------------------------------------------------------------------------------- . . * Manual DiD calculation . di _newline . di "Manual DiD Calculation (Table 1):" Manual DiD Calculation (Table 1): . di "=================================" ================================= . di "Treated change: 96.37 - 60.17 = " %5.2f 96.37 - 60.17 Treated change: 96.37 - 60.17 = 36.20 . di "Control change: 82.10 - 71.22 = " %5.2f 82.10 - 71.22 Control change: 82.10 - 71.22 = 10.88 . di "---------------------------------" --------------------------------- . di "DiD estimate: 36.20 - 10.88 = " %5.2f 36.20 - 10.88 DiD estimate: 36.20 - 10.88 = 25.32 . di _newline . di "The after-school program increased GPA by ~25.32 points." The after-school program increased GPA by ~25.32 points. . di "This is lower than the naive ITS estimate (~36 points)," This is lower than the naive ITS estimate (~36 points), . di "illustrating the importance of using a comparison group." illustrating the importance of using a comparison group. . . . *--------------------------------------------------- . * Section 6: DiD Plots (diff_plot + diff commands) . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 6: DiD PLOTS" SECTION 6: DiD PLOTS . di "========================================" ======================================== . . * Visual DiD plot showing both groups . diff_plot gpa, group(treated) time(post) . graph export "stata_did_diff_plot.png", replace width(2400) (file stata_did_diff_plot.png not found) file stata_did_diff_plot.png written in PNG format . . di "Figure saved: stata_did_diff_plot.png" Figure saved: stata_did_diff_plot.png . . * Formal DiD table using the diff command . diff gpa, treated(treated) period(post) DIFFERENCE-IN-DIFFERENCES ESTIMATION RESULTS -------------------------------------------- Number of observations in the DIFF-IN-DIFF: 70 Before After Control: 25 25 50 Treated: 10 10 20 35 35 -------------------------------------------------------- Outcome var. | gpa | S. Err. | |t| | P>|t| ----------------+---------+---------+---------+--------- Before | | | | Control | 71.215 | | | Treated | 60.166 | | | Diff (T-C) | -11.049 | 0.443 | -24.94 | 0.000*** After | | | | Control | 82.101 | | | Treated | 96.367 | | | Diff (T-C) | 14.266 | 0.443 | 32.20 | 0.000*** | | | | Diff-in-Diff | 25.315 | 0.627 | 40.40 | 0.000*** -------------------------------------------------------- R-square: 0.99 * Means and Standard Errors are estimated by linear regression **Inference: *** p<0.01; ** p<0.05; * p<0.1 . . . *--------------------------------------------------- . * Section 7: DiD Regression Approaches . * Five equivalent methods for estimating the DiD . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 7: DiD REGRESSION APPROACHES" SECTION 7: DiD REGRESSION APPROACHES . di "========================================" ======================================== . . * 7.1 Classical DiD Regression . * Y = alpha + B1*Treat + B2*Post + B3*(Treat x Post) + e . * B3 is the DiD estimate (~25.31) . di _newline . di "--- 7.1 Classical DiD Regression ---" --- 7.1 Classical DiD Regression --- . reg gpa treated post txp, robust Linear regression Number of obs = 70 F(3, 66) = 2660.87 Prob > F = 0.0000 R-squared = 0.9887 Root MSE = 1.184 ------------------------------------------------------------------------------ | Robust gpa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- treated | -11.04936 .2878309 -38.39 0.000 -11.62404 -10.47469 post | 10.88589 .3389564 32.12 0.000 10.20915 11.56264 txp | 25.3149 .6149733 41.16 0.000 24.08706 26.54273 _cons | 71.21514 .2183689 326.12 0.000 70.77915 71.65113 ------------------------------------------------------------------------------ . . * 7.2 Stata Built-in DiD (Stata 17+) . * Note: requires Stata 17+; wrapped in capture for backward compatibility . di _newline . di "--- 7.2 Stata Built-in DiD (didregress, Stata 17+) ---" --- 7.2 Stata Built-in DiD (didregress, Stata 17+) --- . capture noisily didregress (gpa) (txp), group(id) time(time) Treatment and time information Time variable: time Control: txp = 0 Treatment: txp = 1 ----------------------------------- | Control Treatment -------------+--------------------- Group | id | 25 10 -------------+--------------------- Time | Minimum | 1 2 Maximum | 1 2 ----------------------------------- Difference-in-differences regression Number of obs = 70 Data type: Repeated cross-sectional (Std. err. adjusted for 35 clusters in id) ------------------------------------------------------------------------------ | Robust gpa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- ATET | txp | (1 vs 0) | 25.3149 .8337103 30.36 0.000 23.62059 27.0092 ------------------------------------------------------------------------------ Note: ATET estimate adjusted for group effects and time effects. . . * 7.3 Standard Two-Way Fixed Effects (TWFE) with xtreg . * Y = B3*(Treat x Post) + gamma_i + theta_t + e . * Unit FE (gamma_i) absorb time-invariant school differences . * Time FE (theta_t) absorb common shocks . di _newline . di "--- 7.3 Standard TWFE (xtreg) ---" --- 7.3 Standard TWFE (xtreg) --- . xtreg gpa txp i.time, fe vce(cluster id) Fixed-effects (within) regression Number of obs = 70 Group variable: id Number of groups = 35 R-squared: Obs per group: Within = 0.9946 min = 2 Between = 0.4294 avg = 2.0 Overall = 0.8224 max = 2 F(2, 34) = 3360.56 corr(u_i, Xb) = -0.4644 Prob > F = 0.0000 (Std. err. adjusted for 35 clusters in id) ------------------------------------------------------------------------------ | Robust gpa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- txp | 25.3149 .5851062 43.27 0.000 24.12582 26.50398 2.time | 10.88589 .3320367 32.79 0.000 10.21111 11.56067 _cons | 68.05818 .1371096 496.38 0.000 67.77954 68.33682 -------------+---------------------------------------------------------------- sigma_u | 5.1352376 sigma_e | 1.1473933 rho | .9524505 (fraction of variance due to u_i) ------------------------------------------------------------------------------ . . * 7.4 High-Dimensional TWFE with reghdfe . * Faster alternative for models with many fixed effects . di _newline . di "--- 7.4 High-Dimensional TWFE (reghdfe) ---" --- 7.4 High-Dimensional TWFE (reghdfe) --- . reghdfe gpa txp, absorb(id time) cluster(id) (MWFE estimator converged in 2 iterations) HDFE Linear regression Number of obs = 70 Absorbing 2 HDFE groups F( 1, 34) = 1871.90 Statistics robust to heteroskedasticity Prob > F = 0.0000 R-squared = 0.9947 Adj R-squared = 0.9889 Within R-sq. = 0.9814 Number of clusters (id) = 35 Root MSE = 1.1474 (Std. err. adjusted for 35 clusters in id) ------------------------------------------------------------------------------ | Robust gpa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- txp | 25.3149 .5851062 43.27 0.000 24.12582 26.50398 _cons | 73.50113 .0835866 879.34 0.000 73.33126 73.671 ------------------------------------------------------------------------------ Absorbed degrees of freedom: -----------------------------------------------------+ Absorbed FE | Categories - Redundant = Num. Coefs | -------------+---------------------------------------| id | 35 35 0 *| time | 2 1 1 | -----------------------------------------------------+ * = FE nested within cluster; treated as redundant for DoF computation . . * 7.5 TWFE with Covariate (female_share) . * Adding exogenous controls can improve precision . * NOTE: Never control for variables affected by treatment . di _newline . di "--- 7.5 TWFE with Covariate ---" --- 7.5 TWFE with Covariate --- . reghdfe gpa txp female_share, absorb(id time) cluster(id) (MWFE estimator converged in 2 iterations) HDFE Linear regression Number of obs = 70 Absorbing 2 HDFE groups F( 2, 34) = 1125.13 Statistics robust to heteroskedasticity Prob > F = 0.0000 R-squared = 0.9947 Adj R-squared = 0.9886 Within R-sq. = 0.9815 Number of clusters (id) = 35 Root MSE = 1.1609 (Std. err. adjusted for 35 clusters in id) ------------------------------------------------------------------------------ | Robust gpa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- txp | 25.32806 .6047651 41.88 0.000 24.09903 26.55709 female_share | -3.216239 8.700428 -0.37 0.714 -20.89764 14.46516 _cons | 75.19718 4.552635 16.52 0.000 65.94511 84.44925 ------------------------------------------------------------------------------ Absorbed degrees of freedom: -----------------------------------------------------+ Absorbed FE | Categories - Redundant = Num. Coefs | -------------+---------------------------------------| id | 35 35 0 *| time | 2 1 1 | -----------------------------------------------------+ * = FE nested within cluster; treated as redundant for DoF computation . . di _newline . di "All five approaches yield DiD estimate ~25.31-25.33" All five approaches yield DiD estimate ~25.31-25.33 . di "This confirms the manual calculation from Table 1." This confirms the manual calculation from Table 1. . . . *--------------------------------------------------- . * Section 8: Table 2 Replication . * Three specifications exported with outreg2 . * (1) Baseline TWFE . * (2) + Covariate (female_share) . * (3) + Clustered SEs at school level . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 8: TABLE 2 REPLICATION" SECTION 8: TABLE 2 REPLICATION . di "========================================" ======================================== . . * Specification (1): Baseline TWFE, no controls, no clustering . reghdfe gpa i.txp, absorb(id time) (MWFE estimator converged in 2 iterations) HDFE Linear regression Number of obs = 70 Absorbing 2 HDFE groups F( 1, 33) = 1738.48 Prob > F = 0.0000 R-squared = 0.9947 Adj R-squared = 0.9889 Within R-sq. = 0.9814 Root MSE = 1.1474 ------------------------------------------------------------------------------ gpa | Coefficient Std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- 1.txp | 25.3149 .6071435 41.70 0.000 24.07965 26.55014 _cons | 73.50113 .1622659 452.97 0.000 73.171 73.83126 ------------------------------------------------------------------------------ Absorbed degrees of freedom: -----------------------------------------------------+ Absorbed FE | Categories - Redundant = Num. Coefs | -------------+---------------------------------------| id | 35 0 35 | time | 2 1 1 | -----------------------------------------------------+ . outreg2 using table2.doc, replace keep(1.txp) /// > addtext(Controls, No, Clustered SEs, No) dec(2) table2.doc dir : seeout . . * Specification (2): + Covariate (female_share), no clustering . reghdfe gpa i.txp c.female_share, absorb(id time) (MWFE estimator converged in 2 iterations) HDFE Linear regression Number of obs = 70 Absorbing 2 HDFE groups F( 2, 32) = 849.26 Prob > F = 0.0000 R-squared = 0.9947 Adj R-squared = 0.9886 Within R-sq. = 0.9815 Root MSE = 1.1609 ------------------------------------------------------------------------------ gpa | Coefficient Std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- 1.txp | 25.32806 .6148823 41.19 0.000 24.07559 26.58054 female_share | -3.216239 6.607589 -0.49 0.630 -16.67546 10.24298 _cons | 75.19718 3.488308 21.56 0.000 68.09173 82.30263 ------------------------------------------------------------------------------ Absorbed degrees of freedom: -----------------------------------------------------+ Absorbed FE | Categories - Redundant = Num. Coefs | -------------+---------------------------------------| id | 35 0 35 | time | 2 1 1 | -----------------------------------------------------+ . outreg2 using table2.doc, append keep(1.txp) /// > addtext(Controls, Yes, Clustered SEs, No) dec(2) table2.doc dir : seeout . . * Specification (3): No controls, + clustered SEs at school level . reghdfe gpa i.txp, absorb(id time) cluster(id) (MWFE estimator converged in 2 iterations) HDFE Linear regression Number of obs = 70 Absorbing 2 HDFE groups F( 1, 34) = 1871.90 Statistics robust to heteroskedasticity Prob > F = 0.0000 R-squared = 0.9947 Adj R-squared = 0.9889 Within R-sq. = 0.9814 Number of clusters (id) = 35 Root MSE = 1.1474 (Std. err. adjusted for 35 clusters in id) ------------------------------------------------------------------------------ | Robust gpa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- 1.txp | 25.3149 .5851062 43.27 0.000 24.12582 26.50398 _cons | 73.50113 .0835866 879.34 0.000 73.33126 73.671 ------------------------------------------------------------------------------ Absorbed degrees of freedom: -----------------------------------------------------+ Absorbed FE | Categories - Redundant = Num. Coefs | -------------+---------------------------------------| id | 35 35 0 *| time | 2 1 1 | -----------------------------------------------------+ * = FE nested within cluster; treated as redundant for DoF computation . outreg2 using table2.doc, append keep(1.txp) /// > addtext(Controls, No, Clustered SEs, Yes) dec(2) table2.doc dir : seeout . . di _newline . di "Table 2 saved to: table2.doc" Table 2 saved to: table2.doc . di "Expected results:" Expected results: . di " (1) Treatment = 25.31*** (no controls, no clustering)" (1) Treatment = 25.31*** (no controls, no clustering) . di " (2) Treatment = 25.33*** (+ female_share control)" (2) Treatment = 25.33*** (+ female_share control) . di " (3) Treatment = 25.31*** (+ clustered SEs at school level)" (3) Treatment = 25.31*** (+ clustered SEs at school level) . di " All: N=70, R-squared ~0.99" All: N=70, R-squared ~0.99 . di _newline . di "In this simulated example, clustering has minimal effect" In this simulated example, clustering has minimal effect . di "on standard errors. In real-world applications, clustering" on standard errors. In real-world applications, clustering . di "typically changes SEs substantially." typically changes SEs substantially. . . . *=================================================== . * PART 2: EVENT STUDY DESIGN . * Dataset: tutoring_didevent.dta (35 schools x 8 periods) . * Extends the 2x2 DiD to examine dynamic effects . *=================================================== . . . *--------------------------------------------------- . * Section 9: Load and explore the event study dataset . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 9: EVENT STUDY DATA" SECTION 9: EVENT STUDY DATA . di "========================================" ======================================== . . use "https://github.com/quarcs-lab/data-open/raw/master/isds/tutoring_dideven > t.dta", clear . . * Inspect the dataset . describe Contains data from https://github.com/quarcs-lab/data-open/raw/master/isds/tuto > ring_didevent.dta Observations: 280 Variables: 8 13 Apr 2024 11:36 ------------------------------------------------------------------------------- Variable Storage Display Value name type format label Variable label ------------------------------------------------------------------------------- id float %9.0g time float %9.0g treated float %9.0g gpa float %9.0g female_share float %9.0g post float %9.0g txp float %9.0g timeToTreat float %9.0g ------------------------------------------------------------------------------- Sorted by: . summarize Variable | Obs Mean Std. dev. Min Max -------------+--------------------------------------------------------- id | 280 18 10.11759 1 35 time | 280 4.5 2.29539 1 8 treated | 280 .2857143 .4525628 0 1 gpa | 280 80.14277 12.19731 60.07783 107.677 female_share | 280 .521468 .0284086 .4700719 .5698913 -------------+--------------------------------------------------------- post | 280 .5 .5008953 0 1 txp | 280 .1428571 .3505537 0 1 timeToTreat | 80 -.5 2.305744 -4 3 . . * Declare panel structure . xtset id time Panel variable: id (strongly balanced) Time variable: time, 1 to 8 Delta: 1 unit . . * Panel summary . xtsum Variable | Mean Std. dev. Min Max | Observations -----------------+--------------------------------------------+---------------- id overall | 18 10.11759 1 35 | N = 280 between | 10.24695 1 35 | n = 35 within | 0 18 18 | T = 8 | | time overall | 4.5 2.29539 1 8 | N = 280 between | 0 4.5 4.5 | n = 35 within | 2.29539 1 8 | T = 8 | | treated overall | .2857143 .4525628 0 1 | N = 280 between | .4583492 0 1 | n = 35 within | 0 .2857143 .2857143 | T = 8 | | gpa overall | 80.14277 12.19731 60.07783 107.677 | N = 280 between | 1.254617 78.53637 82.74301 | n = 35 within | 12.13424 57.86199 105.5741 | T = 8 | | female~e overall | .521468 .0284086 .4700719 .5698913 | N = 280 between | .0115522 .4933694 .5432372 | n = 35 within | .0260182 .4618352 .5773155 | T = 8 | | post overall | .5 .5008953 0 1 | N = 280 between | 0 .5 .5 | n = 35 within | .5008953 0 1 | T = 8 | | txp overall | .1428571 .3505537 0 1 | N = 280 between | .2291746 0 .5 | n = 35 within | .2677398 -.3571429 .6428571 | T = 8 | | timeTo~t overall | -.5 2.305744 -4 3 | N = 80 between | 0 -.5 -.5 | n = 10 within | 2.305744 -4 3 | T = 8 . . di _newline . di "Panel: 35 schools x 8 time periods = 280 observations" Panel: 35 schools x 8 time periods = 280 observations . di "4 pre-treatment periods + 4 post-treatment periods" 4 pre-treatment periods + 4 post-treatment periods . di "timeToTreat: relative time to treatment onset" timeToTreat: relative time to treatment onset . . . *--------------------------------------------------- . * Section 10: Treatment visualization (panelview) . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 10: EVENT STUDY PANEL VIEW" SECTION 10: EVENT STUDY PANEL VIEW . di "========================================" ======================================== . . panelview gpa txp, i(id) t(time) type(treat) /// > prepost bytiming /// > xtitle("Time Period") ytitle("School ID") /// > legend(position(6)) /// > name(panelview_event, replace) # Variable # Missing % Missing -------------------------------------------- 1 gpa 0 0.0 2 txp 0 0.0 Missing for | how many | variables? | Freq. Percent Cum. ------------+----------------------------------- 0 | 280 100.00 100.00 ------------+----------------------------------- Total | 280 100.00 Note: White cells represent missing values/observations in data. . . graph export "stata_did_panelview_event.png", replace width(2400) (file stata_did_panelview_event.png not found) file stata_did_panelview_event.png written in PNG format . . di "Figure saved: stata_did_panelview_event.png" Figure saved: stata_did_panelview_event.png . . . *--------------------------------------------------- . * Section 11: Event Study Estimation -- Figure 3 . * Replaces single DiD interaction with leads & lags . * Y_it = alpha + sum(theta_j * treat_it(t=k+j)) + gamma_i + theta_t + e . * Leads (pre-treatment): test parallel trends . * Lags (post-treatment): capture dynamic effects . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 11: EVENT STUDY ESTIMATION" SECTION 11: EVENT STUDY ESTIMATION . di " (Figure 3 -- Dynamic Effects)" (Figure 3 -- Dynamic Effects) . di "========================================" ======================================== . . eventdd gpa i.time, timevar(timeToTreat) /// > method(hdfe, absorb(id time) cluster(id)) /// > keepdummies /// > graph_op(ylab(-10(5)30) /// > ytitle("GPA Effect") /// > xtitle("Time to Treatment") /// > xlab(-4(1)4) /// > title("Figure 3: Event Study -- Dynamic Treatment Effects") /// > note("Source: Corral & Yang (2024). Reference period: t = -1.") /// > graphregion(color(white)) plotregion(color(white))) (MWFE estimator converged in 2 iterations) note: 2bn.time is probably collinear with the fixed effects (all partialled-out > values are close to zero; tol = 1.0e-09) note: 3bn.time is probably collinear with the fixed effects (all partialled-out > values are close to zero; tol = 1.0e-09) note: 4bn.time is probably collinear with the fixed effects (all partialled-out > values are close to zero; tol = 1.0e-09) note: 5bn.time is probably collinear with the fixed effects (all partialled-out > values are close to zero; tol = 1.0e-09) note: 6bn.time is probably collinear with the fixed effects (all partialled-out > values are close to zero; tol = 1.0e-09) note: 7bn.time is probably collinear with the fixed effects (all partialled-out > values are close to zero; tol = 1.0e-09) note: 8bn.time is probably collinear with the fixed effects (all partialled-out > values are close to zero; tol = 1.0e-09) HDFE Linear regression Number of obs = 280 Absorbing 2 HDFE groups F( 7, 34) = 1457.71 Statistics robust to heteroskedasticity Prob > F = 0.0000 R-squared = 0.9913 Adj R-squared = 0.9895 Within R-sq. = 0.9610 Number of clusters (id) = 35 Root MSE = 1.2483 (Std. err. adjusted for 35 clusters in id) ------------------------------------------------------------------------------ | Robust gpa | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- time | 2 | 0 (omitted) 3 | 0 (omitted) 4 | 0 (omitted) 5 | 0 (omitted) 6 | 0 (omitted) 7 | 0 (omitted) 8 | 0 (omitted) | lead4 | .3419624 .4012858 0.85 0.400 -.4735485 1.157473 lead3 | -.322034 .4413151 -0.73 0.471 -1.218894 .5748262 lead2 | .593332 .423475 1.40 0.170 -.2672727 1.453937 lag0 | 25.02759 .4450759 56.23 0.000 24.12309 25.93209 lag1 | 24.7052 .5592648 44.17 0.000 23.56863 25.84176 lag2 | 24.76849 .7386192 33.53 0.000 23.26744 26.26955 lag3 | 25.70145 .7965068 32.27 0.000 24.08276 27.32015 _cons | 76.5422 .0735882 1040.14 0.000 76.39265 76.69175 ------------------------------------------------------------------------------ Absorbed degrees of freedom: -----------------------------------------------------+ Absorbed FE | Categories - Redundant = Num. Coefs | -------------+---------------------------------------| id | 35 35 0 *| time | 8 1 7 | -----------------------------------------------------+ * = FE nested within cluster; treated as redundant for DoF computation . . graph export "stata_did_event_study.png", replace width(2400) (file stata_did_event_study.png not found) file stata_did_event_study.png written in PNG format . . di "Figure saved: stata_did_event_study.png" Figure saved: stata_did_event_study.png . . . *--------------------------------------------------- . * Section 12: Table 4 Replication . * Event study coefficients (leads and lags) . *--------------------------------------------------- . . di _newline(2) . di "========================================" ======================================== . di " SECTION 12: TABLE 4 REPLICATION" SECTION 12: TABLE 4 REPLICATION . di "========================================" ======================================== . . outreg2 using table4.doc, replace /// > keep(lead4 lead3 lead2 lag0 lag1 lag2 lag3) dec(2) table4.doc dir : seeout . . di _newline . di "Table 4 saved to: table4.doc" Table 4 saved to: table4.doc . di _newline . di "Event Study Results (Table 4):" Event Study Results (Table 4): . di " Pre-treatment coefficients (leads):" Pre-treatment coefficients (leads): . di " lead4 = " %7.3f _b[lead4] " (SE = " %5.3f _se[lead4] ")" lead4 = 0.342 (SE = 0.401) . di " lead3 = " %7.3f _b[lead3] " (SE = " %5.3f _se[lead3] ")" lead3 = -0.322 (SE = 0.441) . di " lead2 = " %7.3f _b[lead2] " (SE = " %5.3f _se[lead2] ")" lead2 = 0.593 (SE = 0.423) . di " Post-treatment coefficients (lags):" Post-treatment coefficients (lags): . di " lag0 = " %7.3f _b[lag0] " (SE = " %5.3f _se[lag0] ")" lag0 = 25.028 (SE = 0.445) . di " lag1 = " %7.3f _b[lag1] " (SE = " %5.3f _se[lag1] ")" lag1 = 24.705 (SE = 0.559) . di " lag2 = " %7.3f _b[lag2] " (SE = " %5.3f _se[lag2] ")" lag2 = 24.768 (SE = 0.739) . di " lag3 = " %7.3f _b[lag3] " (SE = " %5.3f _se[lag3] ")" lag3 = 25.701 (SE = 0.797) . di _newline . di "Interpretation:" Interpretation: . di " Pre-treatment coefficients are close to zero and mostly" Pre-treatment coefficients are close to zero and mostly . di " insignificant, supporting the parallel trends assumption." insignificant, supporting the parallel trends assumption. . di " Post-treatment coefficients are consistently around 25 points," Post-treatment coefficients are consistently around 25 points, . di " confirming the 2x2 DiD result and showing a constant effect." confirming the 2x2 DiD result and showing a constant effect. . di " N=280, 35 schools, R-squared ~0.992" N=280, 35 schools, R-squared ~0.992 . . . *--------------------------------------------------- . * Section 13: Closing . *--------------------------------------------------- . . di _newline(2) . di "============================================" ============================================ . di " ANALYSIS COMPLETE" ANALYSIS COMPLETE . di "============================================" ============================================ . di " DiD estimate: ~25.32 GPA points" DiD estimate: ~25.32 GPA points . di " Estimand: ATT (Average Treatment on Treated)" Estimand: ATT (Average Treatment on Treated) . di _newline . di " Figures:" Figures: . di " stata_did_panelview_2x2.png" stata_did_panelview_2x2.png . di " stata_did_its.png" stata_did_its.png . di " stata_did_counterfactual.png" stata_did_counterfactual.png . di " stata_did_diff_plot.png" stata_did_diff_plot.png . di " stata_did_panelview_event.png" stata_did_panelview_event.png . di " stata_did_event_study.png" stata_did_event_study.png . di _newline . di " Tables:" Tables: . di " table2.doc (3 regression specifications)" table2.doc (3 regression specifications) . di " table4.doc (event study coefficients)" table4.doc (event study coefficients) . di _newline . di " Reference:" Reference: . di " Corral, D. & Yang, M. (2024). An introduction" Corral, D. & Yang, M. (2024). An introduction . di " to the difference-in-differences design in" to the difference-in-differences design in . di " education policy research." education policy research. . di "============================================" ============================================ . di _newline . di "=== Script completed successfully ===" === Script completed successfully === . . log close name: log: /Users/carlosmendez/Documents/GitHub/starter-academic-v501/content > /post/stata_did/analysis.log log type: text closed on: 26 Apr 2026, 11:26:18 -------------------------------------------------------------------------------