1
$\begingroup$

I have a binomial model with a 3-way interaction:

> all.fit <-  glm(cbind(Presence, Total-Presence) ~ 
                  Season*ToD*Site, family = binomial, 
                  data = all.dt)
> summary(all.fit)

Call:
glm(formula = cbind(Presence, Total - Presence) ~ Season * ToD * 
    Site, family = binomial, data = all.dt)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-6.0379  -1.3245  -0.6221   0.3961   8.6824  

Coefficients:
                                     Estimate Std. Error z value Pr(>|z|)    
(Intercept)                          -4.66074    0.26852 -17.357  < 2e-16 ***
SeasonSpring                          0.60478    0.33133   1.825 0.067959 .  
SeasonSummer                         -0.82682    0.48901  -1.691 0.090873 .  
SeasonWinter                          1.26868    0.30149   4.208 2.58e-05 ***
ToDDay                                0.09865    0.31530   0.313 0.754381    
ToDDusk                               2.48420    0.28183   8.814  < 2e-16 ***
ToDNight                              2.65294    0.27185   9.759  < 2e-16 ***
SiteKawau                             3.09495    0.27605  11.212  < 2e-16 ***
SiteNoises                            3.48048    0.27479  12.666  < 2e-16 ***
SiteTawharanui                        2.94694    0.27684  10.645  < 2e-16 ***
SiteTiritiri                          3.21127    0.27557  11.653  < 2e-16 ***
SeasonSpring:ToDDay                   0.90540    0.37560   2.411 0.015928 *  
SeasonSummer:ToDDay                   2.16685    0.52151   4.155 3.25e-05 ***
SeasonWinter:ToDDay                  -0.12465    0.35416  -0.352 0.724863    
SeasonSpring:ToDDusk                 -2.78750    0.40957  -6.806 1.00e-11 ***
SeasonSummer:ToDDusk                 -1.52114    0.55644  -2.734 0.006263 ** 
SeasonWinter:ToDDusk                 -0.90286    0.32121  -2.811 0.004941 ** 
SeasonSpring:ToDNight                -1.66162    0.34346  -4.838 1.31e-06 ***
SeasonSummer:ToDNight                -1.85826    0.51861  -3.583 0.000339 ***
SeasonWinter:ToDNight                -0.21758    0.30592  -0.711 0.476946    
SeasonSpring:SiteKawau               -0.67217    0.34426  -1.953 0.050875 .  
SeasonSummer:SiteKawau                1.24760    0.49690   2.511 0.012047 *  
SeasonWinter:SiteKawau               -1.68237    0.31709  -5.306 1.12e-07 ***
SeasonSpring:SiteNoises              -0.79023    0.34223  -2.309 0.020940 *  
SeasonSummer:SiteNoises               0.42570    0.49791   0.855 0.392560    
SeasonWinter:SiteNoises              -0.73055    0.31139  -2.346 0.018972 *  
SeasonSpring:SiteTawharanui          -0.25773    0.34405  -0.749 0.453794    
SeasonSummer:SiteTawharanui           1.32456    0.49760   2.662 0.007770 ** 
SeasonWinter:SiteTawharanui          -0.51055    0.31368  -1.628 0.103604    
SeasonSpring:SiteTiritiri            -0.48838    0.34274  -1.425 0.154178    
SeasonSummer:SiteTiritiri             0.93694    0.49749   1.883 0.059656 .  
SeasonWinter:SiteTiritiri            -1.27870    0.31406  -4.072 4.67e-05 ***
ToDDay:SiteKawau                      0.70933    0.32351   2.193 0.028335 *  
ToDDusk:SiteKawau                    -2.35320    0.29549  -7.964 1.67e-15 ***
ToDNight:SiteKawau                   -5.33351    0.29943 -17.812  < 2e-16 ***
ToDDay:SiteNoises                     0.27852    0.32258   0.863 0.387906    
ToDDusk:SiteNoises                   -2.53814    0.29389  -8.636  < 2e-16 ***
ToDNight:SiteNoises                  -4.04432    0.28257 -14.313  < 2e-16 ***
ToDDay:SiteTawharanui                -0.16455    0.32550  -0.506 0.613193    
ToDDusk:SiteTawharanui               -2.76975    0.29927  -9.255  < 2e-16 ***
ToDNight:SiteTawharanui              -2.62701    0.28227  -9.307  < 2e-16 ***
ToDDay:SiteTiritiri                   0.77569    0.32307   2.401 0.016350 *  
ToDDusk:SiteTiritiri                 -2.27149    0.29441  -7.715 1.21e-14 ***
ToDNight:SiteTiritiri                -4.15841    0.28495 -14.593  < 2e-16 ***
SeasonSpring:ToDDay:SiteKawau        -0.19831    0.38943  -0.509 0.610596    
SeasonSummer:ToDDay:SiteKawau        -1.61316    0.53079  -3.039 0.002372 ** 
SeasonWinter:ToDDay:SiteKawau         0.57193    0.37047   1.544 0.122635    
SeasonSpring:ToDDusk:SiteKawau        2.48229    0.43071   5.763 8.25e-09 ***
SeasonSummer:ToDDusk:SiteKawau        1.37786    0.56992   2.418 0.015622 *  
SeasonWinter:ToDDusk:SiteKawau        0.81702    0.34933   2.339 0.019347 *  
SeasonSpring:ToDNight:SiteKawau       1.41405    0.40613   3.482 0.000498 ***
SeasonSummer:ToDNight:SiteKawau       2.06227    0.54597   3.777 0.000159 ***
SeasonWinter:ToDNight:SiteKawau       0.08689    0.36989   0.235 0.814279    
SeasonSpring:ToDDay:SiteNoises       -0.28007    0.38782  -0.722 0.470188    
SeasonSummer:ToDDay:SiteNoises       -1.48112    0.53201  -2.784 0.005369 ** 
SeasonWinter:ToDDay:SiteNoises       -0.06044    0.36566  -0.165 0.868708    
SeasonSpring:ToDDusk:SiteNoises       1.93847    0.43143   4.493 7.02e-06 ***
SeasonSummer:ToDDusk:SiteNoises       0.63232    0.57685   1.096 0.273006    
SeasonWinter:ToDDusk:SiteNoises       0.83278    0.34003   2.449 0.014321 *  
SeasonSpring:ToDNight:SiteNoises      0.69621    0.37365   1.863 0.062422 .  
SeasonSummer:ToDNight:SiteNoises      0.06911    0.55930   0.124 0.901655    
SeasonWinter:ToDNight:SiteNoises      0.48690    0.32194   1.512 0.130428    
SeasonSpring:ToDDay:SiteTawharanui   -0.23150    0.39050  -0.593 0.553287    
SeasonSummer:ToDDay:SiteTawharanui   -1.96624    0.53250  -3.692 0.000222 ***
SeasonWinter:ToDDay:SiteTawharanui    0.43487    0.36870   1.179 0.238216    
SeasonSpring:ToDDusk:SiteTawharanui   2.04614    0.43633   4.689 2.74e-06 ***
SeasonSummer:ToDDusk:SiteTawharanui   0.59597    0.57680   1.033 0.301491    
SeasonWinter:ToDDusk:SiteTawharanui   0.73280    0.34653   2.115 0.034456 *  
SeasonSpring:ToDNight:SiteTawharanui  1.02467    0.36109   2.838 0.004544 ** 
SeasonSummer:ToDNight:SiteTawharanui  1.19957    0.52990   2.264 0.023589 *  
SeasonWinter:ToDNight:SiteTawharanui -0.16571    0.32155  -0.515 0.606306    
SeasonSpring:ToDDay:SiteTiritiri     -0.49278    0.38807  -1.270 0.204143    
SeasonSummer:ToDDay:SiteTiritiri     -1.63113    0.53142  -3.069 0.002145 ** 
SeasonWinter:ToDDay:SiteTiritiri      0.37405    0.36773   1.017 0.309069    
SeasonSpring:ToDDusk:SiteTiritiri     1.81283    0.43042   4.212 2.53e-05 ***
SeasonSummer:ToDDusk:SiteTiritiri     0.83455    0.57225   1.458 0.144744    
SeasonWinter:ToDDusk:SiteTiritiri     0.50070    0.34463   1.453 0.146263    
SeasonSpring:ToDNight:SiteTiritiri    1.49159    0.36761   4.057 4.96e-05 ***
SeasonSummer:ToDNight:SiteTiritiri    1.30689    0.53796   2.429 0.015127 *  
SeasonWinter:ToDNight:SiteTiritiri    0.79478    0.32728   2.428 0.015164 *  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 104046  on 22997  degrees of freedom
Residual deviance:  62740  on 22918  degrees of freedom
AIC: 86778

Number of Fisher Scoring iterations: 6

I have reported the significance of the three way interaction. And now I want to do some comparisons, for example, say Winter at one Site had a higher proportion than Winter at another site. Can I do this following the example here, by changing the level order? (https://rpubs.com/raoulbia/interpreting_glm_logistic_regression_output)

enter image description here

Improve this question
$\endgroup$
6
  • 1
    $\begingroup$ Which events are you counting? This looks like binomial data, so use the format with cbind. If your response is a fraction, some call that a fractional outcome logit see stats.stackexchange.com/questions/530149/… quasibinomial uses quasilikelihood estimation, which does not use directly a binomial likelihood. See stats.stackexchange.com/questions/216122/…, and ... ' $\endgroup$
    – kjetil b halvorsen ♦
    Jun 27, 2021 at 23:51
  • $\begingroup$ ... stats.stackexchange.com/questions/418832/… $\endgroup$
    – kjetil b halvorsen ♦
    Jun 27, 2021 at 23:51
  • $\begingroup$ First make a general test of fhe highest-order interaction, that is, compare a model with interactions of order 3 to a model with only interactions of order 2 $\endgroup$
    – kjetil b halvorsen ♦
    Jun 28, 2021 at 3:11
  • $\begingroup$ Are you talking about choosing the most parsimonious model? Is it possible to answer the question with the existing model? $\endgroup$
    – Louise Wilson
    Jun 28, 2021 at 3:20
  • 1
    $\begingroup$ You can do it, and it is not necessary to change order of levels. What you want to do is test some contrasts, which can be cumbersome in base R when there are interactions. The library on CRAN car have the function linearHypothesis see stats.stackexchange.com/questions/211584/… making this easier, but you will have to read its help page $\endgroup$
    – kjetil b halvorsen ♦
    Jun 28, 2021 at 3:27

1 Answer 1

Reset to default
0
$\begingroup$

You might benefit from the tools L. Hawyard-Smith put together for looking at interactions in count data. https://github.com/lindesaysh/MRSea/blob/master/inst/docs/UsingMRSea_v1.3.pdf

Improve this answer
$\endgroup$

Your Answer

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

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