Gonad Histology Update 3
GLMs are neat
I forgot how useful all of the regression analysis techniques I learned last spring were until I had to use them for my own dataset! I met with Tim last week, who suggested I use a binomial GLM with a logit link to see if treatment affected maturation stage.
I created this R script to analyze my revised dataset.
Any oyster with a maturation stage of 3 (ripe) or 4 (spawning/spent), I considered mature. I then created a new column that assigned 0 for immature and 1 for mature based on the maturation stages. I also renamed
I then used stepwise addition to discern which covariates to put into my model. I started with three separate models to explain maturation: Mature ~ Treatment (ambient vs. low pH), Mature ~ modifiedSex (female vs. male vs. unripe), and Mature ~ Pre.or.Post.OA (pre-treatment vs. post-treatment sampling). The model that used sex to explain maturity was the most significant model, so that became my base. Using
add1, I found that no other variable was significant and needed to be included in the model. Looking at the model summary, I saw that males were typically more mature than females.
For my post-treatment data, I restructured the data to make a contingency table with pH treatment as rows and sex classification (female vs. male vs. unripe) as columns. I then used a poisson GLM with a log link to analyze the contingency table with a chi-squared test for homogeneity. I failed to reject my null hypothesis.
Treatment did not affect maturation. This makes sense with my other results that found egg production did not differ between low and ambient pH treatments. The only phenotype we have so far is lower hatch rates when a cross included a low pH female. I’m really interested to analyze my larval mortality data to see if there is a cohesive narrative.