Dear All,
Recently, I should have been clearer when asking for help interpreting regression coefficients from a high-dimension fixed-effects Poisson regression using the ppmlhdfe command from SSC.
To clarify, I am regressing monthly zip-code-level prescription drug claims on the proportional change in relative annual wellness visits (among other predictors). The "relative annual wellness visits" metric represents the percentage deviation in the rate of wellness visits from a baseline, calculated as:
relative annual wellness visits =(rate of wellness visits at time t/rate of wellness visits at baseline)-1
This measure ranges from -1 (indicating a 100% decline from baseline) to 0.5 (indicating a 50% increase from baseline).
In my regression output, the variable _DDdd1_3_all reflects this metric. Here’s the output for reference (from my older post):
I believe the coefficient of 0.0144727 on _DDdd1_3_all suggests that a 10-percentage-point change in the wellness visit rate would result in a 0.145% change in prescription drug claims, calculated as (exp(0.0144727×0.1)−1)×100.
I'm uncertain about the small percent change implied by this coefficient for our "proportional change" variable. I may be misinterpreting it and would greatly appreciate any insights.
Thank you!
Sumedha
Recently, I should have been clearer when asking for help interpreting regression coefficients from a high-dimension fixed-effects Poisson regression using the ppmlhdfe command from SSC.
To clarify, I am regressing monthly zip-code-level prescription drug claims on the proportional change in relative annual wellness visits (among other predictors). The "relative annual wellness visits" metric represents the percentage deviation in the rate of wellness visits from a baseline, calculated as:
relative annual wellness visits =(rate of wellness visits at time t/rate of wellness visits at baseline)-1
This measure ranges from -1 (indicating a 100% decline from baseline) to 0.5 (indicating a 50% increase from baseline).
In my regression output, the variable _DDdd1_3_all reflects this metric. Here’s the output for reference (from my older post):
Code:
. ppmlhdfe rx_new_cum _DDdd1_3_all unemp_rate deaths if keepc == 1 /// > [w=cellsize], /*eform*/ absorb(i.monthnum i.cohort2 i.zipcode) cluster(zipcohort) (sampling weights assumed) Iteration 1: deviance = 5.0229e+05 eps = . iters = 3 tol = 1.0e-04 min(eta) = -4.10 P Iteration 2: deviance = 4.5978e+05 eps = 9.25e-02 iters = 3 tol = 1.0e-04 min(eta) = -5.50 Iteration 3: deviance = 4.5719e+05 eps = 5.67e-03 iters = 2 tol = 1.0e-04 min(eta) = -6.82 Iteration 4: deviance = 4.5711e+05 eps = 1.82e-04 iters = 2 tol = 1.0e-04 min(eta) = -7.77 Iteration 5: deviance = 4.5710e+05 eps = 7.26e-06 iters = 2 tol = 1.0e-04 min(eta) = -8.43 Iteration 6: deviance = 4.5710e+05 eps = 5.48e-07 iters = 2 tol = 1.0e-05 min(eta) = -8.75 Iteration 7: deviance = 4.5710e+05 eps = 1.22e-08 iters = 2 tol = 1.0e-07 min(eta) = -8.82 S Iteration 8: deviance = 4.5710e+05 eps = 1.29e-11 iters = 2 tol = 1.0e-09 min(eta) = -8.83 S O ------------------------------------------------------------------------------------------------------------ (legend: p: exact partial-out s: exact solver h: step-halving o: epsilon below tolerance) Converged in 8 iterations and 18 HDFE sub-iterations (tol = 1.0e-08) HDFE PPML regression No. of obs = 738,121 Absorbing 3 HDFE groups Residual df = 26,361 Statistics robust to heteroskedasticity Wald chi2(3) = 295.83 Deviance = 457103.9504 Prob > chi2 = 0.0000 Log pseudolikelihood = -3763869.632 Pseudo R2 = 0.0919 Number of clusters (zipcohort)= 26,362 (Std. err. adjusted for 26,362 clusters in zipcohort) --------------------------------------------------------------------------------- | Robust rx_new_cum | Coefficient std. err. z P>|z| [95% conf. interval] ----------------+---------------------------------------------------------------- _DDdd1_3_all | .0144727 .0038331 3.78 0.000 .00696 .0219853 unemp_rate | .0035381 .0017633 2.01 0.045 .000082 .0069942 deaths | 1.07e-06 1.20e-06 0.89 0.371 -1.28e-06 3.42e-06 _cons | -2.417783 .0081682 -296.00 0.000 -2.433793 -2.401774 ---------------------------------------------------------------------------------
I'm uncertain about the small percent change implied by this coefficient for our "proportional change" variable. I may be misinterpreting it and would greatly appreciate any insights.
Thank you!
Sumedha
Comment