# Is this interpretation of mixed ordinal logistic regression correct?

I am doing mixed ordinal logistic regression using clmm function in ordinal package. Before running the clmm model I have changed my DV into ordinal variable using:

> data1.frame$response <- ordered(data1.frame$response)


And this is how my data looks like now:

> str(data1.frame)
'data.frame':   840 obs. of  25 variables:
row         : int  2 13 16 28 34 36 45 46 57 59 ...
response    : Ord.factor w/ 9 levels "1"<"2"<"3"<"4"<..: 8 2 1 2 8 4 8 2 6 2 ...
listeners   : int  300101 300101 300101 300101 300101 300101 300101 300101 300101 300101 ...
group       : Factor w/ 3 levels "english","L2",..: 3 1 3 2 2 2 2 2 2 2 ...
gender      : Factor w/ 2 levels "female","male": 2 2 2 1 1 2 2 1 1 2 ...


Then I run the clmm model:

> mm1 <- clmm (response ~ group*gender + (1|listeners),data=data1.frame)


To see all pair contrasts, I use lsmeans to run pairwise comparison:

> lsmeans(mm1, pairwise~group*gender, adjust="tukey")
$lsmeans group gender lsmean SE df asymp.LCL asymp.UCL english female -1.12179261 0.4774200 NA -2.0576319 -0.1859533 L2 female -0.10179043 0.4502577 NA -0.9843861 0.7808053 thai female -0.32031742 0.4774696 NA -1.2562540 0.6156191 english male -0.17082496 0.4743129 NA -1.1005736 0.7589237 L2 male 0.07489604 0.4506737 NA -0.8085151 0.9583072 thai male -0.46355260 0.4715354 NA -1.3878569 0.4607517 Confidence level used: 0.95$contrasts
contrast                         estimate        SE df    z.ratio p.value
english,female - L2,female    -1.02000218 0.2240555 NA -4.5524528  0.0001
english,female - thai,female  -0.80147519 0.2741795 NA -2.9231766  0.0406
english,female - english,male -0.95096765 0.2691474 NA -3.5332591  0.0055
english,female - L2,male      -1.19668865 0.2258441 NA -5.2987370  <.0001
english,female - thai,male    -0.65824000 0.2631009 NA -2.5018542  0.1235
L2,female - thai,female        0.21852699 0.2243050 NA  0.9742404  0.9262
L2,female - english,male       0.06903453 0.2174003 NA  0.3175456  0.9996
L2,female - L2,male           -0.17668648 0.1593047 NA -1.1091105  0.8778
L2,female - thai,male          0.36176217 0.2111101 NA  1.7136182  0.5227
thai,female - english,male    -0.14949246 0.2691721 NA -0.5553788  0.9938
thai,female - L2,male         -0.39521346 0.2252499 NA -1.7545557  0.4955
thai,female - thai,male        0.14323518 0.2641898 NA  0.5421678  0.9944
english,male - L2,male        -0.24572100 0.2182780 NA -1.1257250  0.8709
english,male - thai,male       0.29272764 0.2585628 NA  1.1321337  0.8681
L2,male - thai,male            0.53844865 0.2124098 NA  2.5349522  0.1141

P value adjustment: tukey method for a family of 6 means


Can I interpret some of the results this way:

1) In female production, rating score of English was significantly lower than L2.

2) In male production, rating score of L2 was higher than Thai although this differences is not significant.

• If you really are primarily interested in within-gender comparisons, the results will be a lot clearer if you use pairwise~group|gender – rvl Dec 18 '14 at 14:07
• Hi rvl, yes the results are clearer, but the p-value stuff is changed. Now I'm not sure which one I should believe. Do you have recommendations on this? – user3288202 Dec 18 '14 at 16:48
• The p values are different because you're focusing on only those 6 comparisons instead of all 15. If you're really not interested in those 9 cross-comparisons that you skipped over, the adjusted p values for the smaller set should be OK. – rvl Dec 19 '14 at 5:15
• That means that if I would keep the same formula, i.e. with all comparisons, it is okay too, isn't it? – user3288202 Dec 19 '14 at 8:17
• Of course. You can surely report all those comparisons. – rvl Dec 19 '14 at 10:37