.

thornoe · Feb 15, 2023 · 23cccd5 · 23cccd5
1 parent 14d56dd
commit 23cccd5
Show file tree

Hide file tree

Showing 28 changed files with 105 additions and 58 deletions.
diff --git a/Example_Stata_LaTeX.do b/Example_Stata_LaTeX.do
@@ -8,8 +8,9 @@
 * MIT LICENSE: Copyright (c) 2023 Thor Donsby Noe (give credit; no liability)
 ********************************************************************************
 /* Installations
-ssc install bcuse 	// access Wooldridge datasets for the examples below
-ssc install estout	// export tables to Excel, Word, or LaTeX
+ssc install bcuse 		// access Wooldridge datasets for the examples below
+ssc install estout		// export tables to Excel, Word, or LaTeX
+ssc install extremes	// list extreme observations for a variable
 */
 
 * Change directory to the folder with your data files (redundant for bcuse)
@@ -60,15 +61,16 @@ label variable trend 	"Time trend"		// covering year t in {0,1,2}
 ********************************************************************************
 * DESCRIPTIVE ANALYSIS (of panel data)
 ********************************************************************************
-* Take a first look at the data
+* Take a first look at the data (by year)
 xtdescribe // balanced panel of 157 firms observed each year 1987-89
-sort year fcode	// sort data by year and firm (required for "by year" command)
-by year: tab grant if scrap!=. // scrap only recorded for 54 of the 157 firms
-by year: sum scrap grant grant_1 if scrap!=. // scrap rate is reduced each year
+correlate scrap year // scrap rate has a negative correlation with time
+correlate scrap grant_lead grant_lead2 if d88==0 & d89==0 // is treatment assignment random?
+bysort year: tab grant if scrap!=. // scrap only recorded for 54 of the 157 firms
 /*	Of the 54 firms for which scrap rate is recorded,
 	19 firms (35.2 %) received a grant in 1988,
 	and another 10 firms (18.5 %) received a grant in 1989.
 */
+bysort year: sum scrap grant grant_1 if scrap!=. // mean scrap rate decreases each year
 
 * Take a closer look at variation in the reduced sample where scrap is observed
 xtsum scrap grant grant_1 if scrap!=. // 54 firms observed over 3 years
@@ -85,13 +87,16 @@ xtsum scrap grant grant_1 if scrap!=. // 54 firms observed over 3 years
      scrap:	Variation between firms > variation over time within each firm
 */
 
+* Identify highest/lowest scrap rates (extreme obs of the first variable listed)
+bysort year: extremes scrap fcode // persistency: the same firms recur each year
+
 * Table: Descriptive statistics - guide: repec.sowi.unibe.ch/stata/estout/esttab.html
 estpost tabstat scrap lscrap grant /// scrap is right skewed (mean right of p50)
 	, statistics(mean sd min p50 max) columns(statistics) /// 
 	listwise // omits obs with any of the chosen variables missing (sample comparable to 'regress')
 esttab using "$tables/descriptive.tex", replace style(tex) delimiter("&") /// create/overwrite LaTeX file
 	label nonumbers nostar postfoot("\hline\end{tabular}") ///
-	cells("mean sd min p50 max") ///
+	cells("mean sd min p50 max") /// format is flexible for each cell
 	stats(N, fmt(%12.0gc) labels("Observations"))
 
 * Table: Descriptive statistics by year
@@ -100,8 +105,8 @@ estpost tabstat scrap lscrap grant ///
 	listwise by(year) // show subsamples by year
 esttab using "$tables/descriptive_yearly.tex", replace style(tex) delimiter("&") ///
 	label nonumbers nostar postfoot("\hline\end{tabular}") ///
-	cells("mean sd min p50 max count") ///
-	noobs // cells("count") makes it redundant to report observations in footer
+	cells("mean sd min p50 max count") /// count for each row; format is flexible
+	noobs // no observations reported in footer as cells("count") reports obs for each row insteads
 
 
 ********************************************************************************
@@ -112,29 +117,35 @@ esttab using "$tables/descriptive_yearly.tex", replace style(tex) delimiter("&")
 gr two	(kdensity scrap if d88==0 & d89==0 & grant_lead==1) ///
 		(kdensity scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0) ///
 		, legend(label(1 "Grant in 1988") label(2 "No grant")) ///
-		title("Panel A: Scrap rates in 1987") xtitle("Scrap rate (per 100 items)") ///
-		ytitle("Density") name(Fig_87, replace) // name for graph combine below
+		title("Panel A: Scrap rates in 1987") ///
+		xtitle("Scrap rate (per 100 items)") ytitle("Density") ///
+		xlab(0(5)30) ylab(0(.05).25) /// fix axis scales to match Fig_89
+		name(Fig_87, replace) // name for graph combine below
 graph export "$figures/kernels_87.png", replace
 sum scrap if d88==0 & d89==0 & grant_lead==1, detail // smallest=.28, p10=.45
 sum scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0, detail // p10=.06
-tab scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0
+tab scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0 // five<.28; three>18
 * Selection bias? No firm getting grant in 1988 scrapped less than 0.28% in 1987
 
 * Figure: Kernel density in 1988 (by grant in 1988)
 gr two	(kdensity scrap if d88==1 & grant==1) ///
 		(kdensity scrap if d88==1 & grant==0 & grant_lead==0) ///
 		, legend(label(1 "Grant in 1988") label(2 "No grant")) ///
-		title("Panel B: Scrap rates in 1988") xtitle("Scrap rate (per 100 items)") ///
-		ytitle("Density") name(Fig_88, replace) // name for graph combine below
+		title("Panel A: Scrap rates in 1988") ///
+		xtitle("Scrap rate (per 100 items)") ytitle("Density") ///
+		xlab(0(5)30) ylab(0(.05).25) /// fix axis scales to match Fig_89
+		name(Fig_88, replace) // name for graph combine below
 graph export "$figures/kernels_88.png", replace
 * A scrap rate < 7% is now more common among the firms that receive grant in 1988
 
 * Figure: Kernel density in 1989 (by grant in 1988)
 gr two	(kdensity scrap if d89==1 & grant_1==1) ///
 		(kdensity scrap if d89==1 & grant_1==0 & grant==0) ///
 		, legend(label(1 "Grant in 1988") label(2 "No grant")) ///
-		title("Panel C: Scrap rates in 1989") xtitle("Scrap rate (per 100 items)") ///
-		ytitle("Density") name(Fig_89, replace) // name for graph combine below
+		title("Panel C: Scrap rates in 1989") ///
+		xtitle("Scrap rate (per 100 items)") ytitle("Density") ///
+		xlab(0(5)30) /// add tick marks for every 5 items scrapped
+		name(Fig_89, replace) // name for graph combine below
 graph export "$figures/kernels_89.png", replace
 * Now a much larger share have a very low scrap rate regardless of grant history
 
@@ -149,15 +160,20 @@ graph export "$figures/kernels_combined.png", replace
 ********************************************************************************
 estimates clear // clear estimates before creating table of estimation results
 
-* Standard pooled OLS as a baseline
+* Baseline (standard pooled OLS with year dummies)
 reg lscrap grant grant_1 d88 d89 // grant is insignificant
 est store baseline, title("Baseline")
 
-* Standard pooled OLS simplified with a yearly time trend instead of year dummies
+* Baseline simplified with a yearly time trend instead of year dummies
 reg lscrap grant grant_1 trend // grant is insignificant
 est store trend, title("Trend")
 
-* Standard pooled OLS with dummies to capture firm-specific effects
+* Take a look at the standard errors (detect outliers)
+predict uhat, residuals // save predicted error term of the last regression
+extremes uhat fcode year scrap, n(10) // unobserved firm-specific effects (permanent differences)
+drop uhat // remove the uhat variable such that it can be predicted again
+
+* Baseline extended with dummies to capture firm-specific effects
 reg lscrap grant grant_1 d88 d89 i.fcode // identical to FE estimation but for constant and dummies
 est store dummies, title("Dummies")
 
@@ -173,13 +189,13 @@ est store FE_cluster, title("FE cluster robust") // see appendix 14A.2 in Wooldr
 estout * using "$tables/results.xls", replace /// create/overwrite Excel workbook
 	starlevels(* .10 ** .05 *** .01) mlabels(,titles numbers) label /// use model titles & variable labels
 	cells( b(star fmt(4)) se(par fmt(4)) ) ///
-	stats( r2 N N_g g_avg, fmt(4 %12.0gc) labels("R-squared" "Obs." "Number of firms" "Obs. per firm") )
+	stats( r2 N N_g g_avg, fmt(%12.4gc) labels("R-squared" "Obs." "Number of firms" "Obs. per firm") )
 
 * Save reduced estimation results as LaTeX file (without dummies)
 estout * using "$tables/results.tex", replace style(tex) /// create/overwrite LaTeX file
 	starlevels(* .10 ** .05 *** .01) mlabels(,titles numbers) label ///
 	cells( b(star fmt(4)) se(par fmt(4)) ) ///
-	stats( r2 N N_g g_avg, fmt(4 %12.0gc) labels("R$^2$" "Obs." "Number of firms" "Obs. per firm") ) ///
+	stats( r2 N N_g g_avg, fmt(%12.4gc) labels("R$^2$" "Obs." "Number of firms" "Obs. per firm") ) ///
 	drop(_cons) indicate("Firm dummies=*fcode*") /// omit constant and firm dummies
 	prehead("\begin{tabular}{lccccc}\hline") /// MANUALLY FIT NUMBER OF C's TO NUMBER OF MODELS!
 	posthead("\hline") prefoot("\hline") ///

diff --git a/Example_Stata_Word.do b/Example_Stata_Word.do
@@ -7,9 +7,10 @@
 * MIT LICENSE: Copyright (c) 2023 Thor Donsby Noe (give credit; no liability)
 ********************************************************************************
 /* Installations
-ssc install bcuse 	// access Wooldridge datasets for the examples below
-ssc install estout	// export tables to Excel, Word, or LaTeX (descriptive)
-ssc install outreg2	// export tables to Excel, Word (estimation results)
+ssc install bcuse 		// access Wooldridge datasets for the examples below
+ssc install estout		// export tables to Excel, Word, or LaTeX (descriptive)
+ssc install outreg2		// export tables to Excel, Word (estimation results)
+ssc install extremes	// list extreme observations for a variable
 */
 
 * Change directory to the folder with your data files (redundant for bcuse)
@@ -60,15 +61,16 @@ label variable trend 	"Time trend"		// covering year t in {0,1,2}
 ********************************************************************************
 * DESCRIPTIVE ANALYSIS (of panel data)
 ********************************************************************************
-* Take a first look at the data
+* Take a first look at the data (by year)
 xtdescribe // balanced panel of 157 firms observed each year 1987-89
-sort year fcode	// sort data by year and firm (required for "by year" command)
-by year: tab grant if scrap!=. // scrap only recorded for 54 of the 157 firms
-by year: sum scrap grant grant_1 if scrap!=. // scrap rate is reduced each year
+correlate scrap year // scrap rate has a negative correlation with time
+correlate scrap grant_lead grant_lead2 if d88==0 & d89==0 // is treatment assignment random?
+bysort year: tab grant if scrap!=. // scrap only recorded for 54 of the 157 firms
 /*	Of the 54 firms for which scrap rate is recorded,
 	19 firms (35.2 %) received a grant in 1988,
 	and another 10 firms (18.5 %) received a grant in 1989.
 */
+bysort year: sum scrap grant grant_1 if scrap!=. // mean scrap rate decreases each year
 
 * Take a closer look at variation in the reduced sample where scrap is observed
 xtsum scrap grant grant_1 if scrap!=. // 54 firms observed over 3 years
@@ -85,23 +87,26 @@ xtsum scrap grant grant_1 if scrap!=. // 54 firms observed over 3 years
      scrap:	Variation between firms > variation over time within each firm
 */
 
+* Identify highest/lowest scrap rates (extreme obs of the first variable listed)
+bysort year: extremes scrap fcode // persistency: the same firms recur each year
+
 * Table: Descriptive statistics - guide: repec.sowi.unibe.ch/stata/estout/esttab.html
 estpost tabstat scrap lscrap grant /// scrap is right skewed (mean right of p50)
 	, statistics(mean sd min p50 max) columns(statistics) /// 
 	listwise // omits obs with any of the chosen variables missing (sample comparable to 'regress')
 esttab using "$tables/descriptive.rtf", replace /// create/overwrite Word document
 	label nonumbers modelwidth(9) ///
-	cells("mean sd min p50 max") ///
+	cells("mean sd min p50 max") /// format is flexible for each cell
 	stats(N, fmt(%12.0gc) labels("Observations"))
 
 * Table: Descriptive statistics by year
 estpost tabstat scrap lscrap grant ///
 	, statistics(mean sd min p50 max count) columns(statistics) /// count obs for each row
 	listwise by(year) // show subsamples by year
 esttab using "$tables/descriptive_yearly.rtf", replace ///
-	label nonumbers modelwidth(8) /// narrow modelwidth requires fewer digits (set manually)
+	label nonumbers modelwidth(8) /// narrow modelwidth only leaves space for 2 digits (set format manually)
 	cells("mean(fmt(2)) sd(fmt(2)) min(fmt(2)) p50(fmt(2)) max(fmt(2)) count(fmt(0))") ///
-	noobs // cells("count") makes it redundant to report observations in footer
+	noobs // no observations reported in footer as cells("count") reports obs for each row instead
 
 
 ********************************************************************************
@@ -112,29 +117,35 @@ esttab using "$tables/descriptive_yearly.rtf", replace ///
 gr two	(kdensity scrap if d88==0 & d89==0 & grant_lead==1) ///
 		(kdensity scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0) ///
 		, legend(label(1 "Grant in 1988") label(2 "No grant")) ///
-		title("Panel A: Scrap rates in 1987") xtitle("Scrap rate (per 100 items)") ///
-		ytitle("Density") name(Fig_87, replace) // name for graph combine below
+		title("Panel A: Scrap rates in 1987") ///
+		xtitle("Scrap rate (per 100 items)") ytitle("Density") ///
+		xlab(0(5)30) ylab(0(.05).25) /// fix axis scales to match Fig_89
+		name(Fig_87, replace) // name for graph combine below
 graph export "$figures/kernels_87.png", replace
 sum scrap if d88==0 & d89==0 & grant_lead==1, detail // smallest=.28, p10=.45
 sum scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0, detail // p10=.06
-tab scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0
+tab scrap if d88==0 & d89==0 & grant_lead==0 & grant_lead2==0 // five<.28; three>18
 * Selection bias? No firm getting grant in 1988 scrapped less than 0.28% in 1987
 
 * Figure: Kernel density in 1988 (by grant in 1988)
 gr two	(kdensity scrap if d88==1 & grant==1) ///
 		(kdensity scrap if d88==1 & grant==0 & grant_lead==0) ///
 		, legend(label(1 "Grant in 1988") label(2 "No grant")) ///
-		title("Panel B: Scrap rates in 1988") xtitle("Scrap rate (per 100 items)") ///
-		ytitle("Density") name(Fig_88, replace) // name for graph combine below
+		title("Panel A: Scrap rates in 1988") ///
+		xtitle("Scrap rate (per 100 items)") ytitle("Density") ///
+		xlab(0(5)30) ylab(0(.05).25) /// fix axis scales to match Fig_89
+		name(Fig_88, replace) // name for graph combine below
 graph export "$figures/kernels_88.png", replace
 * A scrap rate < 7% is now more common among the firms that receive grant in 1988
 
 * Figure: Kernel density in 1989 (by grant in 1988)
 gr two	(kdensity scrap if d89==1 & grant_1==1) ///
 		(kdensity scrap if d89==1 & grant_1==0 & grant==0) ///
 		, legend(label(1 "Grant in 1988") label(2 "No grant")) ///
-		title("Panel C: Scrap rates in 1989") xtitle("Scrap rate (per 100 items)") ///
-		ytitle("Density") name(Fig_89, replace) // name for graph combine below
+		title("Panel C: Scrap rates in 1989") ///
+		xtitle("Scrap rate (per 100 items)") ytitle("Density") ///
+		xlab(0(5)30) /// add tick marks for every 5 items scrapped
+		name(Fig_89, replace) // name for graph combine below
 graph export "$figures/kernels_89.png", replace
 * Now a much larger share have a very low scrap rate regardless of grant history
 
@@ -148,20 +159,25 @@ graph export "$figures/kernels_combined.png", replace
 * ESTIMATION (panel analysis elaborates on Table 14.1 in Wooldridge 7e, p. 464)
 * Replace ".doc" with ".xls" to produce an Excel workbook with results instead
 ********************************************************************************
-* Standard pooled OLS as a baseline
+* Baseline (standard pooled OLS with year dummies)
 reg lscrap grant grant_1 d88 d89 // grant is insignificant
 outreg2 using "$tables/results.doc", replace /// create/overwrite Word document
 	ctitle("Baseline, (se)") label nocons // use variable labels and omit constant
 
-* Standard pooled OLS simplified with a yearly time trend instead of year dummies
+* Baseline simplified with a yearly time trend instead of year dummies
 reg lscrap grant grant_1 trend // grant is insignificant
 outreg2 using "$tables/results.doc", /// append to existing Word table
 	ctitle("Trend, (se)") label nocons
-
-* Standard pooled OLS with dummies to capture firm-specific effects
+
+* Take a look at the standard errors (detect outliers)
+predict uhat, residuals // save predicted error term of the last regression
+extremes uhat fcode year scrap, n(10) // unobserved firm-specific effects (permanent differences)
+drop uhat // remove the uhat variable such that it can be predicted again
+
+* Baseline extended with dummies to capture firm-specific effects
 reg lscrap grant grant_1 d88 d89 i.fcode // identical to FE estimation but for constant and dummies
 outreg2 using "$tables/results.doc", ///
-	drop(i.fcode) addtext(Firm dummies, Yes) /// omit firm dummies
+	drop(i.fcode) addtext(Firm dummies, Yes) /// omit firm dummies but note it
 	ctitle("Dummies, (se)") label nocons
 
 * FE estimation: time-demeaning eliminates firm-specific effect (within-transformation)

diff --git a/Figures/kernels_87.png b/Figures/kernels_87.png
diff --git a/Figures/kernels_88.png b/Figures/kernels_88.png
diff --git a/Figures/kernels_89.png b/Figures/kernels_89.png
diff --git a/Figures/kernels_combined.png b/Figures/kernels_combined.png
diff --git a/LaTeX/bibliography.bib b/LaTeX/bibliography.bib
@@ -1,5 +1,5 @@
 % copy citation from https://Scholar.Google.com
-% or generate using e.g. https://truben.no/latex/bibtex
+% or generate from e.g. https://truben.no/latex/bibtex
 
 @article{holzer1993training,
   title={Are training subsidies for firms effective? The Michigan experience},

diff --git a/LaTeX/figures/kernels_87.png b/LaTeX/figures/kernels_87.png
diff --git a/LaTeX/figures/kernels_88.png b/LaTeX/figures/kernels_88.png
diff --git a/LaTeX/figures/kernels_89.png b/LaTeX/figures/kernels_89.png
diff --git a/LaTeX/figures/kernels_combined.png b/LaTeX/figures/kernels_combined.png
diff --git a/LaTeX/main.pdf b/LaTeX/main.pdf
diff --git a/LaTeX/main.tex b/LaTeX/main.tex
@@ -1,23 +1,23 @@
 \documentclass[11pt]{article}
-\input{preample/packages}
-\input{preample/citations}
-\input{preample/hyperlinks}
-\input{preample/notes}
-\input{preample/ref_commands}
-\input{preample/title_page} % edit title & author(s) in title_page.tex
+\input{preamble/packages}
+\input{preamble/citations}
+\input{preamble/hyperlinks}
+\input{preamble/notes}
+\input{preamble/ref_commands}
+\input{preamble/title_page} % edit title & author(s) in title_page.tex
 \addbibresource{bibliography.bib} % add references to bibliography.bib
 
 %---------------------------------------------------------------
 %-------- TITLE PAGE AND TABLE OF CONTENTS ---------------------
 %---------------------------------------------------------------
 \begin{document}
 \begin{titlingpage}
-    \maketitle
+    \maketitle % using preamble/title_page.tex
     \centering
     \includegraphics[width=.5 \textwidth]{figures/Logo}
     % \begin{abstract}
     % \end{abstract}
-    \vfill % add vertical fill (push contents down to bottom of page)
+    \vfill % adds vertical fill (push contents down to bottom of page)
     \tableofcontents
 \end{titlingpage}
 
@@ -43,7 +43,7 @@
 %---------------------------------------------------------------
 \clearpage
 \appendix
-\setcounter{page}{1}\pagenumbering{roman} % page numbering i, ii, iii, iv...
+\setcounter{page}{1}\pagenumbering{roman} % page numbering i, ii...
 % \renewcommand{\theequation}{\thesection.\arabic{equation}} % add appendix letter to equation numbering
 
 \input{sections/appendix}

diff --git a/LaTeX/preample/citations.tex → LaTeX/preamble/citations.tex b/LaTeX/preample/citations.tex → LaTeX/preamble/citations.tex
diff --git a/LaTeX/preample/hyperlinks.tex → LaTeX/preamble/hyperlinks.tex b/LaTeX/preample/hyperlinks.tex → LaTeX/preamble/hyperlinks.tex
diff --git a/LaTeX/preample/notes.tex → LaTeX/preamble/notes.tex b/LaTeX/preample/notes.tex → LaTeX/preamble/notes.tex
diff --git a/LaTeX/preample/packages.tex → LaTeX/preamble/packages.tex b/LaTeX/preample/packages.tex → LaTeX/preamble/packages.tex
diff --git a/LaTeX/preample/ref_commands.tex → LaTeX/preamble/ref_commands.tex b/LaTeX/preample/ref_commands.tex → LaTeX/preamble/ref_commands.tex
diff --git a/LaTeX/preample/title_page.tex → LaTeX/preamble/title_page.tex b/LaTeX/preample/title_page.tex → LaTeX/preamble/title_page.tex
diff --git a/LaTeX/sections/conclusion.tex b/LaTeX/sections/conclusion.tex
@@ -1,3 +1,6 @@
+% BRIEFLY EXPLAIN WHAT HAS BEEN DONE?
+% BRIEFLY DISCUSS MAIN RESULTS AND CONCLUSIONS
+% BRIEFLY DISCUSS LIMITATIONS AND POSSIBLE EXTENSIONS
 \section{Conclusion}
 \label{s:conclusion}
 The verdict on whether the estimated effects are significant is very sensitive to model choice. Identification of the effects of the training grant is further challenged by a low number of observations, large heterogeneity in scrap rates, and concerns about selection bias.
diff --git a/LaTeX/sections/data.tex b/LaTeX/sections/data.tex
@@ -1,3 +1,6 @@
+% CHARACTERISTICS OF DATASET (SOURCE, TYPE OF DATA)
+% DESCRIPTION OF MAIN VARIABLES
+% DESCRIPTIVE STATISTICS OF MAIN VARIABLES
 \section{Data}
 \label{s:data}
 

diff --git a/LaTeX/sections/discussion.tex b/LaTeX/sections/discussion.tex
@@ -1,3 +1,7 @@
+% ARE RESULTS ROBUST TO DIFFERENT SPECIFICATIONS? (DIFFERENT VARIABLES OR DATA SELECTION)
+% RESULTS' IMPLICATIONS FOR PUBLIC POLICY? (ECONOMIC SIGNIFICANCE)
+% ARE RESULTS COMPARABLE TO THE LITERATURE? (DATA, METHOD, PARAMETER ESTIMATES)
+% POSSIBLE WEAKNESSES AND EXTENSIONS? (BEYOND THE SCOPE OF THIS PAPER)
 \section{Discussion}
 \label{s:discussion}
 

diff --git a/LaTeX/sections/intro.tex b/LaTeX/sections/intro.tex
@@ -15,5 +15,5 @@ \section{Introduction}
 % MAIN RESULTS
 
 
-% OVERVIEW OF THIS PAPER
+% STRUCTURE OF THIS PAPER
 I proceed with presenting the data in \sref{s:data}, empirical model in \sref{s:model}, and results in \sref{s:results} before discussion in \sref{s:discussion} and conclusion in \sref{s:conclusion}.