1
$\begingroup$

So for my current binomial model I am dropping some components and I found out that for one variable the results look a bit different. For 'hurseason' (class factor with two levels Y/N), the LRT is zero. What does this mean for the model and what should I do with it?

Single term deletions

Model:
cr ~ time + dist + habtype + month + hurseason + year + wd + 
    time:dist + dist:habtype + dist:wd + month:wd
             Df Deviance    AIC    LRT  Pr(>Chi)    
<none>            5196.7 5314.7                     
hurseason     0   5196.7 5314.7   0.00              
year          3   5531.5 5643.5 334.76 < 2.2e-16 ***
time:dist     3   5205.7 5317.7   8.95   0.02991 *  
dist:habtype  2   5238.7 5352.7  41.98  7.66e-10 ***
dist:wd       2   5200.9 5314.9   4.23   0.12082    
month:wd     27   5243.2 5307.2  46.52   0.01118 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Additional question: If I drop the dist:wd term (Which is not significant), the AIC goes up by 0.2 compared to the model in which the interaction term is included. I should still drop it right?

EDIT: summary added

Call:
glm(formula = cr ~ time + dist + habtype + month + hurseason + 
    year + wd + time:dist + dist:habtype + month:wd, family = binomial, 
    data = presUVC)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.9229  -0.7871  -0.4491   0.9134   2.9848  

Coefficients: (8 not defined because of singularities)
                 Estimate Std. Error z value Pr(>|z|)    
(Intercept)    -1.619e+01  8.827e+02  -0.018  0.98536    
time11.45      -4.862e-01  1.641e-01  -2.964  0.00304 ** 
time13.45      -1.126e+00  1.957e-01  -5.754 8.72e-09 ***
time19.45      -2.532e+00  1.250e+00  -2.025  0.04286 *  
dist            6.297e-03  1.380e-03   4.563 5.04e-06 ***
habtypeCS      -5.088e-01  3.929e-01  -1.295  0.19530    
habtypeI       -1.518e-01  3.633e-01  -0.418  0.67609    
habtypeP        2.603e+00  3.912e-01   6.655 2.83e-11 ***
month2          3.718e-01  2.500e-01   1.487  0.13691    
month3         -1.910e-08  1.248e+03   0.000  1.00000    
month4          5.936e-01  2.352e-01   2.524  0.01159 *  
month5          1.184e+00  2.312e-01   5.120 3.05e-07 ***
month6          1.342e+00  1.248e+03   0.001  0.99914    
month7          1.257e+00  1.019e+03   0.001  0.99902    
month8          1.432e+00  2.375e-01   6.027 1.67e-09 ***
month9          1.734e+00  1.248e+03   0.001  0.99889    
month10         1.462e+01  8.827e+02   0.017  0.98678    
month11         1.127e+00  2.254e-01   4.999 5.77e-07 ***
month12         8.533e-01  2.270e-01   3.759  0.00017 ***
hurseasonY             NA         NA      NA       NA    
year2013        1.262e-01  1.150e-01   1.098  0.27231    
year2014        1.372e+00  1.106e-01  12.405  < 2e-16 ***
year2015        1.342e+00  1.120e-01  11.990  < 2e-16 ***
wdNW            1.291e+01  8.827e+02   0.015  0.98833    
wdSE            1.340e+01  8.827e+02   0.015  0.98789    
wdSW            1.248e+01  8.827e+02   0.014  0.98872    
time11.45:dist -5.504e-04  4.843e-04  -1.136  0.25579    
time13.45:dist  9.933e-04  5.724e-04   1.735  0.08267 .  
time19.45:dist  4.886e-03  5.057e-03   0.966  0.33403    
dist:habtypeCS -1.024e-06  1.671e-03  -0.001  0.99951    
dist:habtypeI          NA         NA      NA       NA    
dist:habtypeP  -6.304e-03  1.383e-03  -4.559 5.13e-06 ***
month2:wdNW    -2.986e-02  4.197e-01  -0.071  0.94328    
month3:wdNW     6.522e-01  1.248e+03   0.001  0.99958    
month4:wdNW    -2.764e-01  3.942e-01  -0.701  0.48319    
month5:wdNW    -7.382e-01  3.865e-01  -1.910  0.05613 .  
month6:wdNW    -1.106e+00  1.248e+03  -0.001  0.99929    
month7:wdNW    -8.766e-01  1.019e+03  -0.001  0.99931    
month8:wdNW    -5.235e-01  4.115e-01  -1.272  0.20338    
month9:wdNW    -1.231e+00  1.248e+03  -0.001  0.99921    
month10:wdNW   -1.395e+01  8.827e+02  -0.016  0.98739    
month11:wdNW   -4.456e-01  3.797e-01  -1.174  0.24053    
month12:wdNW   -4.344e-01  3.852e-01  -1.128  0.25941    
month2:wdSE     7.377e-01  9.254e-01   0.797  0.42535    
month3:wdSE    -6.140e-01  1.248e+03   0.000  0.99961    
month4:wdSE    -6.456e-01  9.323e-01  -0.693  0.48862    
month5:wdSE    -2.351e+00  1.320e+00  -1.781  0.07488 .  
month6:wdSE    -2.362e+00  1.248e+03  -0.002  0.99849    
month7:wdSE    -1.378e+01  1.041e+03  -0.013  0.98943    
month8:wdSE    -1.088e+00  1.148e+00  -0.947  0.34341    
month9:wdSE    -1.115e+00  1.248e+03  -0.001  0.99929    
month10:wdSE   -1.443e+01  8.827e+02  -0.016  0.98696    
month11:wdSE   -9.748e-01  9.132e-01  -1.067  0.28575    
month12:wdSE   -1.174e+00  1.025e+00  -1.146  0.25195    
month2:wdSW            NA         NA      NA       NA    
month3:wdSW     2.816e-01  1.248e+03   0.000  0.99982    
month4:wdSW            NA         NA      NA       NA    
month5:wdSW            NA         NA      NA       NA    
month6:wdSW    -5.057e-02  1.248e+03   0.000  0.99997    
month7:wdSW    -5.543e-03  1.019e+03   0.000  1.00000    
month8:wdSW            NA         NA      NA       NA    
month9:wdSW    -6.094e-01  1.248e+03   0.000  0.99961    
month10:wdSW   -1.385e+01  8.827e+02  -0.016  0.98749    
month11:wdSW           NA         NA      NA       NA    
month12:wdSW           NA         NA      NA       NA    
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 6499.6  on 5207  degrees of freedom
Residual deviance: 5200.9  on 5151  degrees of freedom
AIC: 5314.9

Number of Fisher Scoring iterations: 13
Improve this question
$\endgroup$
2
  • $\begingroup$ It also has zero degrees of freedom. What does the summary of the model look like? $\endgroup$
    – mdewey
    Jul 20, 2016 at 17:34
  • $\begingroup$ Added the summary of the model to the topic opening post. $\endgroup$
    – Shark167
    Jul 20, 2016 at 18:40

1 Answer 1

Reset to default
1
$\begingroup$

The issue is that the variable was never included in the model in the first place as the summary shows (because of the singularities it warned you about) so when you drop a variable which was never included it has no effect.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thank you. I am pretty new to multiple regression modeling, but what is a good way to deal with singularities (I understand they are perfectly collinear), since I see there are some levels left out (NA) and the hurseason variable, but I want to include the hurseason variable. $\endgroup$
    – Shark167
    Jul 21, 2016 at 7:37
  • $\begingroup$ SOLVED by removing one of the correlated variables. $\endgroup$
    – Shark167
    Jul 21, 2016 at 10:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.