I am using the vglm function in R to compute the effect of my independent variable on my dependent variables (ranging from 0 to 5 or -2 to 2). I also receive outputs for my calculations. However, I don't quite understand how the response variable comparisons are made up. Does the function compare 2 vs. the rest in GenderMale:1 and in GenderMale:2 the 2,1 vs. the rest responses or how can I interpret the output?
Call: vglm(formula = compensate ~ Gender, family = cumulative(parallel = FALSE), data = my_data)
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept):1 -2.7141 0.3113 -8.717 < 2e-16 ***
(Intercept):2 -1.0761 0.1726 -6.234 4.54e-10 ***
(Intercept):3 -0.3307 0.1524 -2.170 0.03002 *
(Intercept):4 1.4733 0.1930 7.634 2.28e-14 ***
GenderMale:1 -1.3406 0.6602 -2.031 0.04230 *
GenderMale:2 -1.0357 0.2980 -3.475 0.00051 ***
GenderMale:3 -0.2784 0.2194 -1.269 0.20441
GenderMale:4 0.1091 0.2783 0.392 0.69499
--- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
IMO the best way is the same way that you may have learned in design of Experiments. There you look at the difference between the the predicted variable values to see what you are estimating with each term in the model.. I learned this in Mendenhall's old DOE now out of print. If you look at the z- library there are other books that probably cover this as well. By Mendenhall.. I would start with the stats for science book. Google all of this for links etc. If you have trouble finding this please contact me on Researchgate. I used to teach this in DOE. KEEP ME UPDATED on your progress please. Good luck with your research, David Booth
BTW I forgot to mention that starts should be given in the package document available on CRAN. Google your starting values question for more details. Best wishes, David Booth
Hi David Eugene Booth ,
Thank you for your tips! I have already googled my question and checked the available documents - without luck so far. But I will try to look at the difference between the predicted variable values, maybe this helps.
Anna are you sure you are using the best R package.? Google ordinal logistic regression in R and check out what that search says. I'm not clear on why you are using vglm. Best wishes, David Booth
I used the polr function (MASS package) for the ordinal logistic regression. However, for some dependent variables the parallel odds assumption does not hold. Therefore, I thought of using the cumulative vglm function (parallel=False) to mitigate this problem. Best regards, Anna
Hi David Eugene Booth ,
I am still stuck with my problem. Do you have any further suggestions where to search for the solution?
Best regards
Anna
Well I am pretty impressed by the MASS book and I knew Brian Ripley before his retirement a number of years ago now. What is the parallel odds assumption. I am not overly knowledgeable but let's see what we can do..Best wishes. David Booth
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R