Abstract
While multiple regression has been a popular statistical choice for postsecondary researchers, it can only tell us the effect of an independent variable on the mean of y. In many applications, however, we would like to know the effect across the entire distribution of y, not just the mean of y. Quantile regression provides one way of telling us this effect, although the interpretation can vary depending upon whether conditional or unconditional quantile regression is used.This chapter reviews conditional and unconditional quantile regression, with an emphasis on the latter as estimated via the recentered influence function, assuming exogeneity of the independent variables, and the instrumental variables quantile treatment effect estimator, assuming endogeneity of treatment. Issues around interpretation, estimation, sensitivity analyses, and presentation of results are discussed, using the 2004 National Survey of Postsecondary Faculty to estimate male-female salary differentials and the effect of faculty unions on faculty salaries.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
This example is based on the discussion at http://www.ats.ucla.edu/stat/stata/faq/quantreg.htm
- 2.
To simplify the analysis, no survey weights or adjustments of the standard errors for the complex sampling design of the NSOPF are used, and the dependent variable is not logged.
- 3.
Indeed, one of the co-authors of the Firpo et al. (2009) paper has done this in their discussion papers, but omitted the sensitivity analyses from their published papers (Fortin, June 2 2014, Personal communication).
- 4.
While their estimator is easily programmed by hand, the ado files for this command can be found at http://faculty.arts.ubc.ca/nfortin/datahead.html
- 5.
Please note that for expository purposes I am assuming selection on observables, but this clearly does not hold here. There are many differences between male and female faculty that are not taken into account by the simple model estimated here, so the results should not be interpreted as the “true” male-female salary differential.
- 6.
Continuous instruments can be dichotomized to satisfy this requirement.
References
Abadie, A. (2003). Semiparametric instrumental variable estimation of treatment response variables. Journal of Econometrics, 113, 231–263.
Abadie, A., Angrist, J., & Imbens, G. (2002). Instrumental variables estimates of the effect of subsidized training on the quantiles of trainee earnings. Econometrica, 70(1), 91–117.
Bielby, R. M., House, E., Flaster, A., & DesJardins, S. L. (2013). Instrumental variables: Conceptual issues and an application considering high school coursetaking. In M. B. Paulsen (Ed.), Higher education: Handbook of theory and Research. New York: Springer.
Buchinsky, M. (1998). Recent advances in quantile regression models: A practical guideline for empirical research. Journal of Human Resources, 33(1), 88–126.
Budig, M. J., & Hodges, M. J. (2010). Differences in disadvantage: Variation in the motherhood penalty across white women’s earnings distribution. American Sociological Review, 75, 705–728.
Cameron, A. C., & Trivedi, P. K. (2005). Microeconometrics: Methods and applications. New York: Cambridge University Press.
Castellano, K. E., & Ho, A. D. (2013a). Contrasting OLS and quantile regression approaches to student “growth” percentiles. Journal of Educational and Behavioral Statistics, 38(2), 190–215.
Castellano, K. E., & Ho, A. D. (2013b). A practioners guide to growth models. Technical report, Council of Chief State School Officers.
Chen, C. L. (2005). An introduction to quantile regression and the quantreg procedure. In SUGI 30 proceedings, Philadelphia.
Cox, N. J. (2007). Kernel estimation as a basic tool for geomorphological data analysis. Earth Surface Processes and Landforms, 32, 1902–1912.
Davino, C., Furno, M., & Vistocco, D. (2014). Quantile regression: Theory and applications. Chichester, UK: Wiley.
Fiorio, C. V. (2004). Confidence intervals for kernel density estimation. The Stata Journal, 4(2), 168–179.
Firpo, S. (2007). Efficient semiparametric estimation of quantile treatment effects. Econometrica, 75(1), 259–276.
Firpo, S., Fortin, N. M., & Lemieux, T. (2009). Unconditional quantile regressions. Econometrica, 77(3), 953–973.
Frandsen, B. R., Fröhlich, M., & Melley, B. (2012). Quantile treatment effects in the regression discontinuity design. Journal of Econometrics, 168, 382–395.
Fröhlich, M., & Melly, B. (2010). Estimation of quantile treatment effects with Stata. The Stata Journal, 10(3), 423–457.
Fröhlich, M., & Melly, B. (2013). Unconditional quantile treatment effects under endogeneity. Journal of Business & Economic Statistics, 31(3), 346–357.
Härdle, W. (1991). Smoothing techniques. New York: Springer.
Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2007). Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Political Analysis, 15(3), 199–236.
Killewald, A., & Bearak, J. (2014). Is the motherhood penalty larger for low-wage women? A comment on quantile regression. American Sociological Review, 79(2), 350–357.
Koenker, R., & Hallock, K. F. (2001). Quantile regression. Journal of Economic Perspectives, 15(4), 143–156.
Maclean, J. C., Webber, D. A., & Marti, J. (2014). An application of unconditional quantile regression to cigarette taxes. Journal of Policy Analysis and Management, 33(1), 188–210.
Mueller, S. (2013). Teacher experience and the class size effect – Experimental evidence. Journal of Public Economics, 98, 44–52.
Murnane, R. J., & Willett, J. B. (2011). Methods matter: Improving causal inference in educational and social science. New York: Oxford University Press.
Porter, S. R. (2013). The causal effect of faculty union on institutional decision-making. Industrial & Labor Relations Review, 66(5), 1192–1211.
Porter, S. R. (2014). Understanding the modern approach to instrumental variables. North Carolina State University. Raleigh: NC.
Salgado-Ugarte, I. H., Shimizu, M., & Taniuchi, T. (1993). Exploring the shape of univariate data using kernel density estimators. Stata Technical Bulletin, 16, 8–19.
Salgado-Ugarte, I. H., Shimizu, M., & Taniuchi, T. (1995). Practical rules for bandwidth selection in univariate density estimation. Stata Technical Bulletin, 27, 5–19.
Scott, D. (1992). Multivariate density estimation. New York: Wiley.
Silverman, B. W. (1992). Density estimation for statistics and data analysis. London: Chapman & Hall.
StataCorp LP (2013). Stata base reference manual release 13. College Station, TX: Stata Press.
Webber, D. A., & Ehrenberg, R. G. (2010). Do expenditures other than instructional expenditures affect graduation and persistence rates in American higher education? Economics of Education Review, 29, 947–958.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendix
Appendix
Below is the Stata syntax used to generate the results in this chapter.
global figures directory
*** Example in Table 8.1 ***
use http://www.ats.ucla.edu/stat/stata/notes/hsb2, clear
sum write, detail
sum write if female==0, detail
sum write if female==1, detail
reg write female
qreg write female
qreg write female, quantile(.25)
replace write=1000 if id==192
reg write female
qreg write female
*** Graphing densities for Fig. 8.6, panel (a) ***
kdensity write, bwidth(1) kernel(gau) legend(off)
graphregion(color(white) lwidth(large)) xtitle
("Writing score") title("")
graph export $figures∖kernel1gau.eps, replace
!epstopdf $figures∖kernel1gau.eps
*** Input faculty salary data ***
use nsopfdata.dta, clear
* Define faculty group for analysis
keep if q1==1 & q2==1 & q3==1 & q5==1 // only instr. duties, faculty status, full-time
keep if q4==1 | q4==2 // principal activity is teaching or research
keep if q10==1 | q10==2 | q10==3 // rank of prof, assoc or asst
* Code independent variables
recode q17a1 (1=1) (0 2/7=0), gen(phd)
recode q71 (2=1) (1=0), gen(female)
gen age=2003-q72
recode q10 (1=1) (2 3=0) (0 4 5 6=.), gen(full)
recode q10 (2=1) (1 3=0) (0 4 5 6=.), gen(assoc)
rename q74b asian
rename q74c black
gen native=0
replace native=1 if q74a==1 | q74d==1
rename q73 latino
rename q52ba articles
rename q52bd books
rename q16cd2 disc
xi i.disc // discipline dummy vars
* Dependent variable
rename q66a basesalary
drop if basesalary<20000 // seems odd to be FT prof and making less than 20K
* Create analytic sample
reg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32
keep if e(sample)
*** OLS-RIF results for Table 8.5 ***
reg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32
estimate store ols
foreach i in 10 25 50 75 90
rifreg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32, quantile(.‘i’)
estimates store q‘i’
estimates table ols q10 q25 q50 q75 q90,
drop(_Idisc_2-_Idisc_32) b(%9.0f) se se(%9.0f)
*** bootstrapping SEs for Table 8.6 ***
foreach i in 10 25 50 75 90
bootstrap, reps(100) seed(642014): rifreg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32, quantile(.‘i’)
estimates store q‘i’
estimates table q10 q25 q50 q75 q90,
drop(_Idisc_2-_Idisc_32) b(%9.0f) se se(%9.0f)
*** Testing sensitivity of results in Table 8.7 ***
* Gaussian
foreach i in 10 25 50 75 90
rifreg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32, quantile(.‘i’)
estimates store silverq‘i’
foreach i in 10 25 50 75 90
rifreg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32, quantile(.‘i’) width(5192)
estimates store hardleq‘i’
foreach i in 10 25 50 75 90
rifreg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32, quantile(.‘i’) width(3802)
estimates store scottq‘i’
estimates table silverq10 silverq25 silverq50 silverq75 silverq90, drop(_Idisc_2-_Idisc_32) b(%9.0f)
se se(%9.0f)
estimates table hardleq10 hardleq25 hardleq50 hardleq75 hardleq90, drop(_Idisc_2-_Idisc_32) b(%9.0f)
se se(%9.0f)
estimates table scottq10 scottq25 scottq50
scottq75 scottq90, drop(_Idisc_2-_Idisc_32) b(%9.0f)
se se(%9.0f)
* to see results with Epanechnikov and uniform
distributions, just add kernop(ep) or kernop(rec) as
options
*** Conditional QR results for Table 8.8 ***
foreach i in 10 25 50 75 90
qreg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32, quantile(.‘i’)
estimates store q‘i’
estimates table q10 q25 q50 q75 q90,
drop(_Idisc_2-_Idisc_32) b(%9.0f) se se(%9.0f)
*** Graph unconditional QR results for gender (Fig. 8.7) ***
* This set of code can be used to create the other figures in the chapter
matrix quantiles = J(1,3,.) // create blank matrix to add model results to
matrix colnames quantiles = B SE Q
matrix identity=J(1,1,1) // to add to counter matrix per loop
matrix counter=J(1,1,0) // will save quatiles for
qraphing
forvalues i=.01(.01)1
matrix counter=counter+identity
qui:rifreg basesalary female asian black latino native full assoc articles books _Idisc_2-_Idisc_32, quantile(‘i’)
matrix table=r(table) // create a matrix of results for each rd (have to rename matrix)
matrix b_se=table[1..2,1..1]’ // grab B and SE and
transpose so they are in column format rather than row
matrix temp=b_se,counter // add quantile as a column
matrix quantiles=quantiles∖temp //add most recent set of model results to matrix
matrix quantiles2=quantiles[2..100,1..3] // drop missing first row
clear svmat quantiles2, names(col) // converts matrix of results to dataset for graphing
gen ciplus=B+1.96*SE
gen cineg=B-1.96*SE
graph twoway connected B Q, msymbol(none) legend(off) graphregion(color(white)) yline(-5540, lpattern (longdash)) lwidth(medthick) xtitle("Quantiles of salary") ytitle(Male-female differential ($)) || connected ciplus Q, msymbol(none) lpattern(dash) || connected cineg Q, msymbol(none) lpattern(dash)
graph export $figures∖gender.eps, replace
!epstopdf "$figures∖gender.eps
*** Finding optimal bandwidths for ivqte command (Table 8.11) ***
locreg facultyunion, logit bandwidth(.2 1.) lambda(.2.5.8) continuous(citi6008) dummy(gov_cons)
locreg facultyunion, logit bandwidth(.05.1.15.2.25) lambda(.05.1.15.2.25) continuous(citi6008) dummy(gov_cons)
locreg facultyunion, logit bandwidth(.06.08.1.12) lambda(.01.02.03.04.05) continuous(citi6008) dummy(gov_cons)
locreg facultyunion, logit bandwidth(.1) lambda(0.0025.005.0075.01) continuous(citi6008) dummy(gov_cons)
*** IV QR estimates for Table 8.12 **
foreach i in 10 25 50 75 90
ivqte basesalary (facultyunion = statelaws), variance quantiles(.‘i’) continuous(citi6008) dummy(gov_cons)
foreach i in 10 25 50 75 90
ivqte basesalary (facultyunion = statelaws), variance quantiles(.‘i’) continuous(citi6008) dummy(gov_cons) bandwidth(.1) lambda(0)
Rights and permissions
Copyright information
© 2015 Springer International Publishing Switzerland
About this chapter
Cite this chapter
Porter, S.R. (2015). Quantile Regression: Analyzing Changes in Distributions Instead of Means. In: Paulsen, M. (eds) Higher Education: Handbook of Theory and Research. Higher Education: Handbook of Theory and Research, vol 30. Springer, Cham. https://doi.org/10.1007/978-3-319-12835-1_8
Download citation
DOI: https://doi.org/10.1007/978-3-319-12835-1_8
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-12834-4
Online ISBN: 978-3-319-12835-1
eBook Packages: Humanities, Social Sciences and LawEducation (R0)