Difference in IRR for same variable in poisson

[Mithila shrestja]

Hello everyone,
When I use the command : poisson glioma_cases i.sex, exposure(pop) irr, I obtained IRR for male 1.248(95% CI 1.02-1.29)
Then I again repeat poisson glioma_cases i.AgeGroup i.sex, exposure(pop) irr. Now, IRR for male is different ie. 1.33(95% CI 1.27-1.37).
I am wondering why it is different.
Any help would be appreciated.
Thank you.

[Richard Williams]

If you change the model the coefficients will usually change at least a little bit (sometimes a lot). It is also possible that the sample changed a bit too.

FYI, it is much easier to comment if you include the actual code and output using code tags. See pt. 12 of the FAQ.

[Carlo Lazzaro]

Mithila:
is -i.AgeGroup- in your second model that makes the difference?

[Mithila shrestja]

Hello Carlo,
No, it is similar for i.AgeGroup

[Carlo Lazzaro]

Mithila:
I meant that, as per your post, it seems that -i.AgeGroup- was not included among your 1st model predictors.

[Mithila shrestja]

Is it making the difference?
But when look for i.AgeGroup alone as poisson glioma_cases i.AgeGroup, exposure(pop) irr and again as poisson glioma_cases i.AgeGroup i.sex, exposure(pop) irr. The results are similar for i. AgeGroup.

[Carlo Lazzaro]

Mithila:
as per Richard's suggestion (that echoes a FAQ), please provide the list with what you typed and what Stata gave you back. Thanks.

[Mithila shrestja]

Stata output looks like this with command poisson glioma_cases i.sex, exposure(pop) irr

Iteration 0: log likelihood = -5676.9229
Iteration 1: log likelihood = -5676.906
Iteration 2: log likelihood = -5676.906

Poisson regression Number of obs = 1584
LR chi2(1) = 129.25
Prob > chi2 = 0.0000
Log likelihood = -5676.906 Pseudo R2 = 0.0113


glioma_cases IRR Std. Err. z P>z [95% Conf. Interval]

2.sex 1.24828 .0243999 11.35 0.000 1.201362 1.297031
_cons .0000425 6.12e-07 -699.49 0.000 .0000414 .0000438

ln(pop) 1 (exposure)




Then, I repeat as
poisson glioma_cases i.AgeGroup i.sex, exposure(pop) irr

Iteration 0: log likelihood = -3742.0483
Iteration 1: log likelihood = -3724.2132
Iteration 2: log likelihood = -3724.117
Iteration 3: log likelihood = -3724.117

Poisson regression Number of obs = 1584
LR chi2(18) = 4034.83
Prob > chi2 = 0.0000
Log likelihood = -3724.117 Pseudo R2 = 0.3514

------------------------------------------------------------------------------
glioma_cases | IRR Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
AgeGroup |
2 | .9394085 .0816199 -0.72 0.472 .7923159 1.113809
3 | .7610622 .069181 -3.00 0.003 .636862 .9094836
4 | .8854272 .0764898 -1.41 0.159 .7475147 1.048784
5 | .9470672 .0798423 -0.65 0.519 .8028241 1.117226
6 | 1.583218 .1195471 6.08 0.000 1.365423 1.835752
7 | 2.06967 .1499456 10.04 0.000 1.795694 2.385447
8 | 2.171547 .1566763 10.75 0.000 1.885191 2.5014
9 | 2.354508 .1683461 11.98 0.000 2.046631 2.708698
10 | 2.938874 .2050536 15.45 0.000 2.563246 3.369549
11 | 3.632972 .2492939 18.80 0.000 3.175797 4.15596

|
2.sex | 1.325147 .026037 14.33 0.000 1.275085 1.377174
_cons | .000017 1.05e-06 -177.09 0.000 .000015 .0000191
ln(pop) | 1 (exposure)

We could notice IRR for sex is different from the previous one.

[Clyde Schechter]

This is perfectly normal, expected behavior in regression models. The first model is giving you a crude estimate of the sex IRR; the second is giving you an estimate adjusted (coarsely perhaps) for age. There is no reason to expect the results to be the same.

[Mithila shrestja]

Thank you. I have queries again.
I have time series data in following format for years 1970-2013.
dg_y AgeGroup sex pop glioma_cases
1970 0-4 female 166273 5
1970 5-9 female 186729 5
1970 10-14 female 194975 4
1970 15-19 female 205481 1
1970 20-24 female 216751 5

dg_y AgeGroup sex pop glioma_cases
1971 0-4 female 160437 2
1971 5-9 female 184936 6
1971 10-14 female 191684 2
1971 15-19 female 206166 2
1971 20-24 female 214936 0

dg_y AgeGroup sex pop glioma_cases
1970 0-4 male 173171 5
1970 5-9 male 194431 6
1970 10-14 male 202971 3
1970 15-19 male 215689 3
1970 20-24 male 228348 4

dg_y AgeGroup sex pop glioma_cases
1971 0-4 male 167275 1
1971 5-9 male 192445 4
1971 10-14 male 199232 4
1971 15-19 male 215563 1
1971 20-24 male 227202 2


Now, I would like to calculate age adjusted incidence rates for 1970-2013.

I used :
poisson glioma_cases i.dg_y i.AgeGroup, ir

Iteration 0: log likelihood = -3882.3336
Iteration 1: log likelihood = -3872.3088
Iteration 2: log likelihood = -3872.1944
Iteration 3: log likelihood = -3872.1943

Poisson regression Number of obs = 1584
LR chi2(60) = 4490.95
Prob > chi2 = 0.0000
Log likelihood = -3872.1943 Pseudo R2 = 0.3670

------------------------------------------------------------------------------
glioma_cases | IRR Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
dg_y |
1971 | .9565217 .1164529 -0.37 0.715 .7534671 1.214298
1972 | 1.065217 .126259 0.53 0.594 .8443962 1.343786
1973 | 1.021739 .1223469 0.18 0.857 .8080048 1.292011
1974 | 1.050725 .1249564 0.42 0.677 .832263 1.326531
1975 | 1.043478 .1243046 0.36 0.721 .8261974 1.317902
1976 | 1.275362 .1450115 2.14 0.032 1.020587 1.593739
1977 | 1.036232 .1236524 0.30 0.766 .8201325 1.309272
1978 | 1.463768 .1616577 3.45 0.001 1.17887 1.817517
1979 | 1.384058 .1546308 2.91 0.004 1.111874 1.722871
1980 | 1.57971 .1718441 4.20 0.000 1.276386 1.955117


it takes 1970 as reference data here and others are relative risk with 1970 as reference year.

I want age adjusted incidence rates for all years from 1970-2013 not the incidence rate ratio.

I should use it in joint-point regression to see distinct change in trends and to calculate annual percentage change.

[Maarten Buis]

You can use the ibn. prefix instead of the i. prefix and add the nocons option. See: http://maartenbuis.nl/publications/ref_cat.html

[Mithila shrestja]

Thank you Maarten,
I used ibn.prefix as
poisson glioma_cases ibn.dg_y ibn.AgeGroup, noconstant
note: 18.AgeGroup omitted because of collinearity

Iteration 0: log likelihood = -4712.7613
Iteration 1: log likelihood = -3879.0241
Iteration 2: log likelihood = -3872.1993
Iteration 3: log likelihood = -3872.1943
Iteration 4: log likelihood = -3872.1943

Poisson regression Number of obs = 1584
Wald chi2(61) = 50379.35
Log likelihood = -3872.1943 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
glioma_cases | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
dg_y |
1970 | -.9515735 .1552439 -6.13 0.000 -1.255846 -.6473009
1971 | -.9960252 .1563012 -6.37 0.000 -1.30237 -.6896805
1972 | -.8883946 .1538084 -5.78 0.000 -1.189853 -.5869356
1973 | -.9300673 .1547466 -6.01 0.000 -1.233365 -.6267696
1974 | -.9020934 .1541131 -5.85 0.000 -1.20415 -.6000372


I got coefficient not adjusted rates. Is it possible to findout adjusted rates in someway?

[Maarten Buis]

You still need to specify the irr option.

[Richard Williams]

As an aside, please use code tags when posting output. It makes things much easier to read. See pt 12 of the FAQ

[Clyde Schechter]

In response to #12, adding the -irr- option will give you the incidence rate ratios for each year, relative to whatever is your reference (omitted) year. To get the adjusted rates themselves in each year, you can run the -margins- command with the -predict(ir)- option after -poisson-:

margins dg_y, predict(ir)