Strange output for glmmTMB and pairwise comparison

Question

I am running a glmmTMB to see if there is a significant difference in survival to the eyed egg stage (proportional data between 0 and 1) depending on what genetic male type was used (W, YY, or F1) to fertilize said eggs. I have female fish as my random effect as each female fish is a repeated measure. Below is my code and output:

model_glmm_Eyed <- glmmTMB(Eyed ~ Genetic + 
        (1 | FemaleID), data = df, 
        family = beta_family())

Conditional model:
        Estimate Std. Error z value Pr(>|z|)    
(Intercept)   1.4800     0.1991   7.432 1.07e-13 ***
GeneticW      0.3309     0.2936   1.127    0.260    
GeneticYY    -0.4053     0.2472  -1.640    0.101   

emmeans(model_glmm_Eyed, pairwise ~ Genetic)

$emmeans
Genetic emmean    SE  df asymp.LCL asymp.UCL
F1        1.48 0.199 Inf     1.090      1.87
W         1.81 0.269 Inf     1.284      2.34
YY        1.07 0.203 Inf     0.676      1.47

Results are given on the logit (not the response) scale. 
Confidence level used: 0.95 

 $contrasts
contrast estimate    SE  df z.ratio p.value
F1 - W     -0.331 0.294 Inf  -1.127  0.4972
F1 - YY     0.405 0.247 Inf   1.640  0.2290
W - YY      0.736 0.299 Inf   2.460  0.0370

Results are given on the log odds ratio (not the response) scale. 
P value adjustment: tukey method for comparing a family of 3 estimates 

I have two questions regarding this analysis

1. Why am I not seeing results for males with genetic type F1? Are they getting absorbed in the intercept and being used as the baseline comparison? If so, how do I change this?

2. If I run post-hoc test, it shows me the pairwise comparison for all three male types. It shows a significant difference in the survival to the eyed stage between W and YY male types. I am confused about how to interpret this. When I visually look at my data it appears that there should be a difference in the survival to the eyed stage between W and YY males but it isn't appearing that way in the glmmTMB.

Any insight into why I might be getting these weird results or if I should run a different test would be greatly appreciated! :)

Answer 1

Welcome to CrossValidated.

For your first question, that is a question about the regression coefficients, and it reflects how the model is parameterized. With one factor and the default parameterization, the intercept is the estimate for group F1, and the other coefficients are what gets added to the intercept to get the other two means. Note that this matches the emmeans estimate: 1.48, 1.48 + 0.33 = 1.81, and 1.48 - 0.41 = 1.07.

For the second question, that's how it works with statistical tests. You have an estimated effect, and there is uncertainty in that estimate. Unless the effect rises above the uncertainty, it is not "significant." In this particular example, there is the additional confusion that the estimates are on a log-odds-ratio scale, which probably is not the scale that you visualized them on.