Interpreting xtpoisson fixed effects with reference category

[Charles Bondi]

Here is my code and abbreviated output. The dependent variable is total payments for inpatient services. The first three independent variables are health plan types: consumer directed, comprehensive, and hmo-similar.

I have two years of data, with 0-1 indicators for each year and person. So I have a reference group, ppo. I want to be sure I'm interpreting the estimated coefficient for hmo-similar correctly, especially in regard to the reference group.

My interpretation: When a person changes from a different plan type to hmo-similar, the dependent variable changes by 14.1%. But this leaves out the idea of relationship to the reference group. Should my interpretation be "changes by 14.1% more than if a person moves to a ppo?" Or am I missing something here?

I've checked Cameron & Trivedi and other sources and there doesn't seem to be an example that clears this up.

Thanks for any help or insights. - Charles Bondi

. xtpoisson totpay_I cdhp comp hmo_sim style_msa1 pract year eeclass_* eestatu_* state_* if (le99person_any&stateok&nocap_person&e xpok_person&drgdaysok), fe vce(robust)
note: you are responsible for interpretation of non-count dep. variable
note: 14542 groups (14542 obs) dropped because of only one obs per group
note: 770369 groups (1540738 obs) dropped because of all zero outcomes
note: eeclass_09 omitted because of collinearity
note: eestatu_09 omitted because of collinearity
note: state_31 omitted because of collinearity
note: state_65 omitted because of collinearity
note: state_09 dropped because it is constant within group
note: state_27 dropped because it is constant within group
note: state_59 dropped because it is constant within group
note: state_63 dropped because it is constant within group

Iteration 0: log pseudolikelihood = -1.562e+09
Iteration 1: log pseudolikelihood = -1.552e+09
Iteration 2: log pseudolikelihood = -1.552e+09
Iteration 3: log pseudolikelihood = -1.552e+09
Iteration 4: log pseudolikelihood = -1.552e+09
Iteration 5: log pseudolikelihood = -1.552e+09
Iteration 6: log pseudolikelihood = -1.552e+09
Iteration 7: log pseudolikelihood = -1.552e+09
Iteration 8: log pseudolikelihood = -1.552e+09
Iteration 9: log pseudolikelihood = -1.552e+09
Iteration 10: log pseudolikelihood = -1.552e+09
Iteration 11: log pseudolikelihood = -1.552e+09
Iteration 12: log pseudolikelihood = -1.552e+09
Iteration 13: log pseudolikelihood = -1.552e+09
Iteration 14: log pseudolikelihood = -1.552e+09

Conditional fixed-effects Poisson regression Number of obs = 257,538
Group variable: enrolid Number of groups = 128,769

Obs per group:
min = 2
avg = 2.0
max = 2

Wald chi2(67) = 928.27
Log pseudolikelihood = -1.552e+09 Prob > chi2 = 0.0000

(Std. Err. adjusted for clustering on enrolid)
------------------------------------------------------------------------------
| Robust
totpay_I | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
cdhp | .0214846 .1203678 0.18 0.858 -.214432 .2574012
comp | .0574204 .0576624 1.00 0.319 -.0555958 .1704366
hmo_sim | .1411482 .0826132 1.71 0.088 -.0207707 .3030671
style_msa1| .0017563 .0008885 1.98 0.048 .0000149 .0034978
practpercap| -.7698588 1.554977 -0.50 0.621 -3.817559 2.277841
year | .0077341 .0261489 0.30 0.767 -.0435168 .058985

[Charles Bondi]

Separate estimations that alternate the reference group do answer my own question. The implied estimation of the reference group is zero, as the absolute differences are always .141.