Clustering in Fixed Effect Regression with Reshaped Panel Data

[Andrew Clarke]

Hi,

I am working on a fixed effects regression analysis using a panel dataset with monthly data on location units and price changes, influenced by variables that vary across different countries. My dataset initially had the country-level variable in a wide format, with each country represented as a separate column such that I ran this regression:

foreach country in africa middleeast  {
    xtreg dspread_`country' ldspread_`country' L(1/12).unct_`country', fe robust
}

To improve my analysis I would like to do country year clustering to account for location cross time fixed effects and the cross sectional correlation and time series dependence. The regression I am trying to run would look like this:

xtreg dspread_* ldspread_* L(1/12).unct_*, fe vce(cluster country yearvar)

Therefore, I reshaped the data into a long format to incorporate the country variable more effectively. However, post-reshaping, each unique location-time combination (ward_num and month) now corresponds to multiple rows due to the added country variable. In other words for each location month, there is a row for "africa" and one for "middleeast". This causes issues when trying to set my panel data for fixed effects regression:

reshape long spread_ wui_ dspread_ ldspread_ dwui_ unct_, i(ward_num monthlyvar) j(country) string
gen yearvar = year(dofm(monthlyvar))
sort ward_num monthlyvar

. xtset ward_num monthlyvar
repeated time values within panel
r(451);

Could anyone suggest a workaround or alternative approach for this data structure issue? How can I effectively incorporate country-year clustering in this setup?

Thank you in advance for your help and insights!


The data looks like this after reshaping:

* Example generated by -dataex-. For more info, type help dataex
clear
input long ward_num float(monthlyvar yearvar) str14 country float(dspread_ ldspread_ unct_)
1 660 2015 "africa" -.01273128 .014274824 0
1 660 2015 "middleeast" -.017508611 .019569613 0
1 661 2015 "africa" -.015691198 -.01273128 0
1 661 2015 "middleeast" .03139296 -.017508611 0
1 662 2015 "africa" .08255005 -.015691198 0
1 662 2015 "middleeast" -.034069523 .03139296 0
1 663 2015 "africa" -.05759017 .08255005 0
1 663 2015 "middleeast" .016335353 -.034069523 0
1 664 2015 "africa" .035960466 -.05759017 0
1 664 2015 "middleeast" -.008077189 .016335353 0
1 665 2015 "africa" .016428001 .035960466 0
1 665 2015 "middleeast" -.031927027 -.008077189 0
1 666 2015 "africa" -.0044620577 .016428001 0
1 666 2015 "middleeast" -.008583674 -.031927027 0
1 667 2015 "africa" -.0003889892 -.0044620577 0
1 667 2015 "middleeast" -.02492235 -.008583674 0
1 668 2015 "africa" -.024398034 -.0003889892 0
1 668 2015 "middleeast" .05400065 -.02492235 0
1 669 2015 "africa" .02820484 -.024398034 1
1 669 2015 "middleeast" -.001000002 .05400065 0
1 670 2015 "africa" -.01822365 .02820484 0
1 670 2015 "middleeast" -.008926474 -.001000002 0
1 671 2015 "africa" .0039288327 -.01822365 0
1 671 2015 "middleeast" .01623942 -.008926474 0
1 672 2016 "africa" -.002127517 .0039288327 0
1 672 2016 "middleeast" -.01431036 .01623942 0
1 673 2016 "africa" .04208383 -.002127517 0
1 673 2016 "middleeast" .0015582517 -.01431036 0
1 674 2016 "africa" -.01246523 .04208383 0
1 674 2016 "middleeast" -.04449745 .0015582517 0
1 675 2016 "africa" .05236107 -.01246523 0
1 675 2016 "middleeast" .01941245 -.04449745 0
1 676 2016 "africa" -.026499826 .05236107 0
1 676 2016 "middleeast" -.01979743 .01941245 0
1 677 2016 "africa" -.02334593 -.026499826 0
1 677 2016 "middleeast" -.02490872 -.01979743 0
1 678 2016 "africa" .015093558 -.02334593 1
1 678 2016 "middleeast" -.03613212 -.02490872 1
1 679 2016 "africa" -.011342783 .015093558 0
1 679 2016 "middleeast" .025140826 -.03613212 0
1 680 2016 "africa" .04714107 -.011342783 0
1 680 2016 "middleeast" .033584423 .025140826 0
1 681 2016 "africa" -.07134539 .04714107 0
1 681 2016 "middleeast" -.011004753 .033584423 0
1 682 2016 "africa" .05198849 -.07134539 0
1 682 2016 "middleeast" .005849187 -.011004753 0
1 683 2016 "africa" -.02892708 .05198849 1
1 683 2016 "middleeast" .013701008 .005849187 0
2 660 2015 "africa" -.01273128 .014274824 0
2 660 2015 "middleeast" -.017508611 .019569613 0
2 661 2015 "africa" -.015691198 -.01273128 0
2 661 2015 "middleeast" .03139296 -.017508611 0
2 662 2015 "africa" .08255005 -.015691198 0
2 662 2015 "middleeast" -.034069523 .03139296 0
2 663 2015 "africa" -.05759017 .08255005 0
2 663 2015 "middleeast" .016335353 -.034069523 0
2 664 2015 "africa" .035960466 -.05759017 0
2 664 2015 "middleeast" -.008077189 .016335353 0
2 665 2015 "africa" .016428001 .035960466 0
2 665 2015 "middleeast" -.031927027 -.008077189 0
2 666 2015 "africa" -.0044620577 .016428001 0
2 666 2015 "middleeast" -.008583674 -.031927027 0
2 667 2015 "africa" -.0003889892 -.0044620577 0
2 667 2015 "middleeast" -.02492235 -.008583674 0
2 668 2015 "africa" -.024398034 -.0003889892 0
2 668 2015 "middleeast" .05400065 -.02492235 0
2 669 2015 "africa" .02820484 -.024398034 1
2 669 2015 "middleeast" -.001000002 .05400065 0
2 670 2015 "africa" -.01822365 .02820484 0
2 670 2015 "middleeast" -.008926474 -.001000002 0
2 671 2015 "africa" .0039288327 -.01822365 0
2 671 2015 "middleeast" .01623942 -.008926474 0
2 672 2016 "africa" -.002127517 .0039288327 0
2 672 2016 "middleeast" -.01431036 .01623942 0
2 673 2016 "africa" .04208383 -.002127517 0
2 673 2016 "middleeast" .0015582517 -.01431036 0
2 674 2016 "africa" -.01246523 .04208383 0
2 674 2016 "middleeast" -.04449745 .0015582517 0
2 675 2016 "africa" .05236107 -.01246523 0
2 675 2016 "middleeast" .01941245 -.04449745 0
2 676 2016 "africa" -.026499826 .05236107 0
2 676 2016 "middleeast" -.01979743 .01941245 0
2 677 2016 "africa" -.02334593 -.026499826 0
2 677 2016 "middleeast" -.02490872 -.01979743 0
2 678 2016 "africa" .015093558 -.02334593 1
2 678 2016 "middleeast" -.03613212 -.02490872 1
2 679 2016 "africa" -.011342783 .015093558 0
2 679 2016 "middleeast" .025140826 -.03613212 0
2 680 2016 "africa" .04714107 -.011342783 0
2 680 2016 "middleeast" .033584423 .025140826 0
2 681 2016 "africa" -.07134539 .04714107 0
2 681 2016 "middleeast" -.011004753 .033584423 0
2 682 2016 "africa" .05198849 -.07134539 0
2 682 2016 "middleeast" .005849187 -.011004753 0
2 683 2016 "africa" -.02892708 .05198849 1
2 683 2016 "middleeast" .013701008 .005849187 0
3 660 2015 "africa" -.01273128 .014274824 0
3 660 2015 "middleeast" -.017508611 .019569613 0
3 661 2015 "africa" -.015691198 -.01273128 0
3 661 2015 "middleeast" .03139296 -.017508611 0
end
format %tm monthlyvar
label values ward_num ward_num
label def ward_num 1 "E05009317", modify
label def ward_num 2 "E05009318", modify
label def ward_num 3 "E05009319", modify

[Andrew Clarke]

Forgot to ask whether this would also work in this context?

reghdfe dspread_* ldspread_* L(1/12).unct_*, absorb(ward_num monthlyvar) vce(cluster country yearvar)

This command produced warnings about the VCV matrix being non-positive semi-definite and the necessity of an adjustment. Additionally, I get no F-statistic.

. reghdfe dspread_* ldspread_* L(1/12).unct_*, absorb(ward_num monthlyvar) vce(cluster country yearvar)
(MWFE estimator converged in 5 iterations)
note: L.unct_ is probably collinear with the fixed effects (all partialled-out values are close to zero; tol = 1.0e-09)
Warning: VCV matrix was non-positive semi-definite; adjustment from Cameron, Gelbach & Miller applied.
warning: missing F statistic; dropped variables due to collinearity or too few clusters

HDFE Linear regression Number of obs = 1,033,620
Absorbing 2 HDFE groups F( 12, 9) = .
Statistics robust to heteroskedasticity Prob > F = .
R-squared = 0.3939
Adj R-squared = 0.3935
Number of clusters (country) = 14 Within R-sq. = 0.2108
Number of clusters (yearvar) = 10 Root MSE = 0.0286

(Std. err. adjusted for 10 clusters in country yearvar)