Ancova

mia

New Member
#1
hei,

i am comparing body-measyrments of two groups of fish that are fed different diets against each other. the problem is that the two diets lead two very different growt; one grow quickly and the other slowly. thus, the variance of body length (one of the predictor variables) are unequal unless i log-tramsform. the other predictor variable is type of diet. the response variable are other body measurments, i.e. jaw and eye length. when i do levene's test on the response variable i get a significant difference between the two diets (after trasnformation). however, when i plot the residuals agaisnt the fitted values (after transfoming both the response and the predictor variable) the variance seem okay.

so my question is; do i have to perfom a leven's test to show that the variance is equal or can i refer to the plotting of the model when i say that the data are not violating the assumptions of ANCOVA.

thank you so much for ur answer

Regards, mia
 
#2
You're saying that after transforming the variables (log), your Levene's is fine? Then why not use the transformed measurements?

-e

hei,

i am comparing body-measyrments of two groups of fish that are fed different diets against each other. the problem is that the two diets lead two very different growt; one grow quickly and the other slowly. thus, the variance of body length (one of the predictor variables) are unequal unless i log-tramsform. the other predictor variable is type of diet. the response variable are other body measurments, i.e. jaw and eye length. when i do levene's test on the response variable i get a significant difference between the two diets (after trasnformation). however, when i plot the residuals agaisnt the fitted values (after transfoming both the response and the predictor variable) the variance seem okay.

so my question is; do i have to perfom a leven's test to show that the variance is equal or can i refer to the plotting of the model when i say that the data are not violating the assumptions of ANCOVA.

thank you so much for ur answer

Regards, mia
 

mia

New Member
#3
no my levene's test is not fine. that is my problem; transfroming the response variable doesn't make the levene's okay. but when i transform both the response and the continouos explanatory variables the variance seem okay when i visually "check" it by plotting the residuals against the fitted values.

my qusetion is then: in order to use ANCOVA do i have to say that the levene's test is not significant or is it okay to say that i visually checked it?

thanx for all help,
mia