Interpretation of IRR values in percentage of NBRM

[Md Ohid]

I have a data-set which contains members profile of two parliaments (eight and nine) and their questions counts . Here I use dummy variable nineparl where nine parliament=1, else=0. and dependent variable totques and another dummy variable opposall where members from opposition=1, else =0. Certainly I have many other variables which I am not stated here to shorten the question. Here is NBRM IRR results in percentage of unit change:

Variables	IRR	Impact of One unit Change
nineparl	1.440	44%
opposall	1.113	11%

Now my question is how do I interpret this value?

[Carlo Lazzaro]

Md:
welcome to the list.
I would guess the following temptative answer (asuming that youìve replaced -poisson- with -nbreg- due to a real overdispersion issue detected after having run -poisson-):
-when contrasted against the reference category (majority of eight parliament), being member of opposal increase the questions count of about 11%, whereas being member of the nine parliament increase that value of about 44%.
That said, your model misses the interaction term between the two categorical variables (see -help fvvarlist- and related entry in Stata .pdf manual about that topic).
For the future, please post what you tyoed and what Stata gave you back via CODE delimiters (as reminded by FAQ). Thanks.

[Marcos Almeida]

Just a side note, obvious to most but, anyway, something I wish to underline, only point estimates (without p-values and CIs) for the predictors were showed, hence some of them may (or may not) be statistically significant.

Better yet, since this was your first post in this Forum, I want to recall (as we can read in the FAQ advice) that the best way to interpret the model is showing the whole output "exactly".

[Md Ohid]

.Dear Marcos,

Thanks for your suggestions. Here I am posting my dataset example below:

 dataex totques nineparl opposall in 1/15

copy starting from the next line ------ ----------------

* Example generated by -dataex-. To install: ssc install    dataex
clear
input byte totques float(nineparl opposall)
0 1 0
12 1 0
0 1 0
20 1 1
0 1 0
0 1 0
0 1 0
0 1 0
6 1 1
0 1 0
5 1 0
0 1 0
0 1 1
0 1 0
0 1 1
end

copy up to and including the previous line - ----------------

Listed 15 out of 311 observations

I am using

nbreg nineparl opposall

first then

nbreg,irr

Now I am requesting you please suggest me to explain my irr values in percentage terms.

[Carlo Lazzaro]

Md:
unfrtunately you excerpt does not seem to give back any informative outcome (or the one that you expected):

. input byte totques float(nineparl opposall)

      totques   nineparl   opposall
  1.
. 0 1 0
  2.
. 12 1 0
  3.
. 0 1 0
  4.
. 20 1 1
  5.
. 0 1 0
  6.
. 0 1 0
  7.
. 0 1 0
  8.
. 0 1 0
  9.
. 6 1 1
 10.
. 0 1 0
 11.
. 5 1 0
 12.
. 0 1 0
 13.
. 0 1 1
 14.
. 0 1 0
 15.
. 0 1 1
 16.
. end

. nbreg nineparl opposall

Fitting Poisson model:

Iteration 0:   log likelihood =        -15 
Iteration 1:   log likelihood =        -15 

Fitting constant-only model:

Iteration 0:   log likelihood = -20.794415 
Iteration 1:   log likelihood =        -15 
Iteration 2:   log likelihood =        -15 

Fitting full model:

Iteration 0:   log likelihood =        -15 
Iteration 1:   log likelihood =        -15 

Negative binomial regression                    Number of obs     =         15
                                                LR chi2(1)        =       0.00
Dispersion     = mean                           Prob > chi2       =     1.0000
Log likelihood =        -15                     Pseudo R2         =     0.0000

------------------------------------------------------------------------------
    nineparl |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    opposall |          0   .5838742     0.00   1.000    -1.144372    1.144372
       _cons |          0   .3015113     0.00   1.000    -.5909514    .5909514
-------------+----------------------------------------------------------------
    /lnalpha |   -26.8093          .                             .           .
-------------+----------------------------------------------------------------
       alpha |   2.27e-12          .                             .           .
------------------------------------------------------------------------------
LR test of alpha=0: chibar2(01) = 0.00                 Prob >= chibar2 = 1.000

. nbreg,irr

Negative binomial regression                    Number of obs     =         15
                                                LR chi2(1)        =       0.00
Dispersion     = mean                           Prob > chi2       =     1.0000
Log likelihood =        -15                     Pseudo R2         =     0.0000

------------------------------------------------------------------------------
    nineparl |        IRR   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    opposall |          1   .5838742     0.00   1.000     .3184237     3.14047
       _cons |          1   .3015113     0.00   1.000     .5538002    1.805706
-------------+----------------------------------------------------------------
    /lnalpha |   -26.8093          .                             .           .
-------------+----------------------------------------------------------------
       alpha |   2.27e-12          .                             .           .
------------------------------------------------------------------------------
LR test of alpha=0: chibar2(01) = 0.00                 Prob >= chibar2 = 1.000

.

[Md Ohid]

Dear Carlo,

Here I posted output of my NBRM:

 nbreg,irr

Negative binomial regression                    Number of obs     =        615
                                                LR chi2(5)        =      20.39
Dispersion     = mean                           Prob > chi2       =     0.0011
Log likelihood = -1435.5217                     Pseudo R2         =     0.0071

------------------------------------------------------------------------------
     totques |        IRR   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     elecvul |   1.080468   .2139873     0.39   0.696     .7328763    1.592916
    opposall |   .8633803   .1974664    -0.64   0.521     .5514691    1.351709
    business |   1.290644   .2635603     1.25   0.212     .8649363    1.925879
    staffcap |   .4236608   .1137052    -3.20   0.001      .250361    .7169188
    nineparl |   1.718216   .3481815     2.67   0.008     1.155019    2.556033
       _cons |   1.048657   .2143022     0.23   0.816     .7025596    1.565251
ln(duration) |          1  (exposure)
-------------+----------------------------------------------------------------
    /lnalpha |   1.724948   .0775618                      1.572929    1.876966
-------------+----------------------------------------------------------------
       alpha |   5.612228   .4352947                       4.82075    6.533653
------------------------------------------------------------------------------
LR test of alpha=0: chibar2(01) = 7229.52              Prob >= chibar2 = 0.000

Here is the descriptive statistics of above variables:

 sum totques  elecvul opposall business staffcap nineparl duration

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
     totques |        615    6.946341    12.34686          0         79
     elecvul |        615    .5056911    .5003746          0          1
    opposall |        615    .2520325    .4345333          0          1
    business |        615    .5382114    .4989436          0          1
    staffcap |        615    .1691057    .3751504          0          1
-------------+---------------------------------------------------------
    nineparl |        615    .5056911    .5003746          0          1
    duration |        615     4.50993    1.204284   .0876112   5.015743

You can see all independent variables are dummy variables. Here elecvul=1 when win margin is upto 30%, else=0. All opposition is 1, else=0. MPs from business profession is 1 others zero. MPs who have staff capacity is 1, other zero. Ninth parliament is one, eight parliament=0 in our dataset only eight and nine parliament's questions count are recorded. duration is computed in years. It is regarded as exposure variable.

So I believe, you can now interpret the IRR values based on the above table. I am using Satata SE 14.2.

[Carlo Lazzaro]

Md:
provided that it's up to you (not me) to interpret the results of your regression (my contribution could help you out only) and limiting the issue to statistical significant coefficinets only:
-Other things being equal, (being component of) ninth parliament increases the -totques- rate of 1.72 times;
-Other things being equal, having a staff reduces the -totques- rate of (1-.4236608)=.5763 times.

[Marcos Almeida]

First things first. If by interpretation you mean understanding the ordinary output of a standard nbreg model, there are an infinite number of places we can find information, starting by the Stata Manual. Here, another good source:https://stats.idre.ucla.edu/stata/ou...al-regression/

[Md Ohid]

Dear Marcos and Carlo,

Thanks for your responses. I already go through the link you provided before I posted my questions here. But I am not fully understand their interpretation to customize my case. From my previous post you have seen nineparl in positively related and statistically significant for the case of total questions asked in both eight and nine parliament. But we consider the total supplementary questions instead of total questions this value was negative and also significant. Here I post the output:

Negative binomial regression                    Number of obs     =        616
                                                LR chi2(5)        =      65.41
Dispersion     = mean                           Prob > chi2       =     0.0000
Log likelihood = -1183.0299                     Pseudo R2         =     0.0269

------------------------------------------------------------------------------
    totsuppl |        IRR   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     elecvul |    1.06701   .1339078     0.52   0.605     .8343426     1.36456
    opposall |   .5090441   .0785715    -4.37   0.000     .3761592    .6888731
    business |   .8272568   .1039836    -1.51   0.131     .6466168    1.058361
    staffcap |   .5031538   .0872679    -3.96   0.000     .3581515    .7068622
    nineparl |    .438151   .0567976    -6.37   0.000     .3398462    .5648917
       _cons |   1.055463   .1520198     0.37   0.708     .7958725    1.399725
ln(duration) |          1  (exposure)
-------------+----------------------------------------------------------------
    /lnalpha |    .581863   .0875264                      .4103144    .7534116
-------------+----------------------------------------------------------------
       alpha |   1.789369    .156617                      1.507292    2.124235
------------------------------------------------------------------------------
LR test of alpha=0: chibar2(01) = 980.68               Prob >= chibar2 = 0.000

You certainly see the IRR value for nineparl is (1-0.438)=0.562. Then what I say here, like: Other things being equal, number of members or number of female members in the ninth parliament decreases the number of supplementary questions by 0.562 times. But actually female members are more in number in ninth parliament than eight parliament even they asked more supplementary questions than their counterpart in eight. Therefore I am confused how to interpret this irr values.

Please clarify me if my understanding is wrong.

I have also find another form of interpretation, such as:

For every one-standard deviation increases in ninth parliament the rate of number of supplementary questions decrease by nearly 56 percent , holding all else equal.

Does it worked for me?

Thanks,
Ohid

[Marcos Almeida]

Sorry, but your model is not clear to me.

If you didn't select a variable to specify the sex of individuals, why you say that the "number of female members in the ninth parliament decreases the number of supplementary question by 0.562 times"?

[Carlo Lazzaro]

Md:
I do share Marcos' concerns about your model specification.

[Md Ohid]

Sorry Carlos and Marcos,

My main model has female member but I wrongly mentioned it. Actually it should be number of businessmen or member of opposition in the model which I posted above. My dependable variable is number of supplementary questions in the above model but it was number of total questions in the previous model. I hope it will clear now.

[Marcos Almeida]

Please, read the FAQ, particularly this advice:

12.1 What to say about your commands and your problem

Say exactly what you typed and exactly what Stata typed (or did) in response. N.B. exactly!

That being said, the interpretation of the IRR is "exactly" the one which has been discussed so far.

However, it assumes the model is not misspecified.

As Carlo appropriatetly underlined in #7, this task is up to you to be in charge.