Stats

7-13-11 Homegeneity testing 750
Tried the logistic regression 2 ways. Using link = "log" in my glm for the 750 data was having difficulty converging. Reading up on this, it seems like this is a problem with this type of regression. No problems using the logit function although it seems that this is more applicable for normally distributed, linear data.

Talking with BV - reporting the variance of all trials is a good idea due to the inherent level of noise in the data due to differences in temp, dose, larvae robustness, etc.

Procedure: GLM.22 <- glm(p ~ Trial + pCO2 + Dose, family=binomial(logit), data=week750, weights=n)

> GLM.33 <- glm(p ~ Trial + pCO2 + Dose, family=binomial(logit), data=d, weights=n)

> summary(GLM.33)

RESULTS: Either way I got a significant difference between experiments using the 750ppm.

6-21-11 - tests of homogeneity between experiments - LD50
Goal: LD50 data at 2000ppm - to combine or not to combine?

Procedure: week<-read.csv(file.choose("LD50_2000_all_week_old.csv")) head(week)

week$Day <- as.factor(week$Day) week$Dose<-as.factor(week$Dose) week$p<-(week$Live/week$n) week$y<-cbind(week$Dead,week$n-week$Dead) GLM.1 <- glm(p ~ Trial + pCO2 + Dose, family=binomial(link="log"), data=week, weights=n) summary(GLM.1)

RESULTS: Week old larvae: Trials A (4/27) and B (5/20) can be combined but not C (6/3) or D (6/23) is significantly different (C: p=0.00126; D:~0). Next step - plot to look at the differences. D-hinge larvae: Trials B and D can be combined, but not C (B+D: p= 0.779)