I am working on the model that influences the counts of climate change legislation in state legislatures. As expected, I decided to use the zinb model because there are so many excess zeros.
When I employed this syntax, my rationale was that contributions from two types of interest groups to the elections (fossil_campaign2 and climate_campaign2) might also exhibit excess zeros. That was the resaon for including these two variables within the inflate() function. However, after understanding additional manuals and explanations, I think this inflate function actually refers to the possible variables that account for the group of "excess zeros" in the dependent variables, irrespective of whether the independent variables have excess zeros themselves.
With this clarification in mind, I have a couple of questions:
1, Is that right?
2. How should I interpret the negative coefficients for the two variables? Given that it's a logit link function (Unsure), does it imply that an increase of one unit in fossil fuel campaign contributions decreases the likelihood of being in the zero-inflated group for the dependent variable? Is the effect size too strong?
3. How can I understand an insignificant intercept in this model?
4. From a theoretical standpoint, individual ideology significantly influences climate change legislation (Usually Conservative legislators would not introduce climate change legislation, so could be the main factor that influences excess zeros). With this in mind, should I adjust the syntax as follows?
Many thanks
Code:
zinb incentive_count c.std_candper c.ideoscore c.tenure2 i.session, irr nolog vce (cluster sfips) inflate(c.fossil_campaign2 c.climate_campaign2)
Zero-inflated negative binomial regression Number of obs = 1,644
Inflation model: logit Nonzero obs = 495
Zero obs = 1,149
Wald chi2(8) = .
Log pseudolikelihood = -1599.069 Prob > chi2 = .
(Std. err. adjusted for 6 clusters in sfips)
-----------------------------------------------------------------------------------
| Robust
incentive_count | IRR std. err. z P>|z| [95% conf. interval]
------------------+----------------------------------------------------------------
incentive_count |
std_candper | 1.148491 .1433607 1.11 0.267 .8992416 1.466828
ideoscore | .3068753 .1067686 -3.40 0.001 .1551709 .6068948
tenure2 | .991072 .009914 -0.90 0.370 .9718301 1.010695
|
session |
2 | .7671407 .1665842 -1.22 0.222 .50123 1.174121
3 | .6766692 .0922604 -2.86 0.004 .5179877 .8839615
4 | .702565 .1297458 -1.91 0.056 .4892055 1.008978
5 | .4691583 .2901897 -1.22 0.221 .1395801 1.57694
6 | .3404974 .2996782 -1.22 0.221 .0606674 1.911052
|
_cons | 1.955293 1.083214 1.21 0.226 .660168 5.791211
------------------+----------------------------------------------------------------
inflate |
fossil_campaign2 | -4.742035 1.143721 -4.15 0.000 -6.983686 -2.500383
climate_campaign2 | .7751751 1.894358 0.41 0.682 -2.937698 4.488048
_cons | .0018064 .5451719 0.00 0.997 -1.066711 1.070324
------------------+----------------------------------------------------------------
/lnalpha | -.2441682 .3563587 -0.69 0.493 -.9426184 .4542821
------------------+----------------------------------------------------------------
alpha | .7833559 .2791557 .3896064 1.575042
With this clarification in mind, I have a couple of questions:
1, Is that right?
2. How should I interpret the negative coefficients for the two variables? Given that it's a logit link function (Unsure), does it imply that an increase of one unit in fossil fuel campaign contributions decreases the likelihood of being in the zero-inflated group for the dependent variable? Is the effect size too strong?
3. How can I understand an insignificant intercept in this model?
4. From a theoretical standpoint, individual ideology significantly influences climate change legislation (Usually Conservative legislators would not introduce climate change legislation, so could be the main factor that influences excess zeros). With this in mind, should I adjust the syntax as follows?
Code:
zinb incentive_count c.fossil_campaign2 c.climate_campaign2 c.std_candper c.tenure2 i.session, irr nolog vce (cluster sfips) inflate(c.ideoscore)
Zero-inflated negative binomial regression Number of obs = 1,644
Inflation model: logit Nonzero obs = 495
Zero obs = 1,149
Wald chi2(9) = .
Log pseudolikelihood = -1628.714 Prob > chi2 = .
(Std. err. adjusted for 6 clusters in sfips)
-----------------------------------------------------------------------------------
| Robust
incentive_count | IRR std. err. z P>|z| [95% conf. interval]
------------------+----------------------------------------------------------------
incentive_count |
fossil_campaign2 | 1.179792 .1097568 1.78 0.076 .9831448 1.415771
climate_campaign2 | .846464 .1036474 -1.36 0.173 .665857 1.076059
std_candper | 1.149116 .0996086 1.60 0.109 .9695702 1.36191
tenure2 | 1.003058 .0121582 0.25 0.801 .9795092 1.027173
|
session |
2 | .8798232 .2128926 -0.53 0.597 .5475559 1.413717
3 | .5319398 .0707656 -4.74 0.000 .4098495 .6903995
4 | .6535677 .1069918 -2.60 0.009 .4741823 .9008156
5 | .3270673 .1911053 -1.91 0.056 .1040592 1.028002
6 | .2498645 .19415 -1.78 0.074 .0544883 1.145793
|
_cons | 1.366394 1.347718 0.32 0.752 .197701 9.44372
------------------+----------------------------------------------------------------
inflate |
ideoscore | 1.057072 .5515854 1.92 0.055 -.0240154 2.13816
_cons | -.9381537 .5600537 -1.68 0.094 -2.035839 .1595314
------------------+----------------------------------------------------------------
/lnalpha | -.1012308 .9857309 -0.10 0.918 -2.033228 1.830766
------------------+----------------------------------------------------------------
alpha | .9037244 .890829 .1309123 6.238665
