xtpoisson prediction

[amera amery]

Hi,

I am using xtpoisson fixed effect regression. After getting the coefficients as follows, I want to make the prediction and get the accuracy of the model. I see simple predict gave me numbers in specific range for event count.

Any help please?

. xtpoisson count average_temprature average_precipitation humidity  avg_shifted commulitave_2_years, fe irr
note: 1038 groups (21798 obs) dropped because of all zero outcomes
note: humidity dropped because it is constant within group

Iteration 0:   log likelihood = -35735.979  
Iteration 1:   log likelihood = -35032.504  
Iteration 2:   log likelihood = -35032.009  
Iteration 3:   log likelihood = -35032.009  

Conditional fixed-effects Poisson regression        Number of obs    =  41,643
Group variable: fips                                Number of groups =   1,983

                                                    Obs per group:
                                                                 min =      21
                                                                 avg =    21.0
                                                                 max =      21

                                                    Wald chi2(4)     = 1394.43
Log likelihood = -35032.009                         Prob > chi2      =  0.0000

---------------------------------------------------------------------------------------
                count |        IRR   Std. err.      z    P>|z|     [95% conf. interval]
----------------------+----------------------------------------------------------------
   average_temprature |   1.208188   .0065024    35.14   0.000      1.19551    1.220999
average_precipitation |   .9954726   .0108513    -0.42   0.677     .9744301     1.01697
          avg_shifted |   1.005804   .0009224     6.31   0.000     1.003998    1.007613
  commulitave_2_years |   1.000069   .0001511     0.46   0.646     .9997733    1.000366
---------------------------------------------------------------------------------------

[Jared Greathouse]

What do we mean "get the accuracy of the model"?

You also say

I see simple predict gave me numbers in specific range for event count.

implying that you've used the predict post-estimation command. I don't understand what the issue is here.

[amera amery]

I meant getting predicted values (y hate) and comparing it with real values (Y) to check the accuracy. how to get y hate values?

[Rich Goldstein]

see

help xtpoisson postestimation##predict

[amera amery]

I found this code in one post and it gives accurate results. But why do we have to go through this to get y hat for Poisson regression ? Any references?

 
 sysuse auto, clear g id=rep78 drop if id==. xtset id xtpoisson price mpg, fe predict double fitted, xb gen double yhat=exp(fitted)  egen meany=mean(price) , by(id)  egen meanyhat=mean(yhat), by(id)  gen double exp_alpha=meany/meanyhat if meanyhat>0  replace yhat=yhat*exp_alpha if meanyhat>0 su price yhat

[Jared Greathouse]

I still don't understand.

I'm not at my computer, but if we did

xtpoisson y x

predict yhat

we get our predictions from the model, no additional complications needed. As Rich said, go to your stata terminal and type

help xtpoisson postestimation##predict

[Joao Santos Silva]

Dear amera amery et al.:

I am afraid it is not possible to obtain meaningful predictions after xtpoisson with FE; that is illustrated here.

However, in your case you have a reasonable T and therefore so you can get predictions by adapting the code below

clear all
sysuse auto
xtset rep78
xtpoisson price mpg , fe
predict double xb if e(sample)==1
g double yhat=exp(xb)
egen meany=mean(price) if e(sample)==1, by(rep78)
egen meanyhat=mean(yhat) , by(rep78)
gen double exp_alpha=meany/meanyhat
replace yhat=yhat*exp_alpha

Best wishes,

Joao

[amera amery]

@Joao Santos Silva Thank you very much, this code works with my model and gives very close values for the dependent variable. But what I meant is why we have to do this process, why predict command does not give the direct predicted value?

[Joao Santos Silva]

Dear amera amery

That is because to get the fitted values you need the fixed effects and most fixed effects estimators wipe them out rather than estimating them. A good textbook should explain that.

Best wishes,

Joao