# Effect size in linear mixed models

Thank you for reading this question.

I know there have been a few discussions regarding this topic, but I couldn't get a satisfactory answer. So here is my question, with some details in the beginning-

I am trying to understand an effect of a chemical on insect's body mass. So I have 3 groups - untreated insects, insects fed with dose 1, and insects fed with dose 2. Multiple experimental blocks. I am measuring body mass to see an effect of the chemical. I am running a linear mixed model in R, fixed effect = treatment, random effect = block, family = gaussian.

I my linear model, I am comparing "untreated" with "treated dose 1 and dose 2" As an output I get values for these:

Intercept Estimate Std.Error DF t-value p-value

What I gather from other blogs - the "Estimate" value is where you get your effect size from. Is that correct?

If estimate value is 0.01, I need to convert it to odds ratio by exp(0.01) exp(0.01) is = 1 roughly.

This means, that there is no effect of my chemical treatment, correct?

I will also look at p value and confidence intervals along with odds ratio, but is my overall understanding fine?

Is there any other way of getting effect sizes in linear mixed models?

Thank you!

• Why would you want to use an odds ratio as your estimated effect when your response variable is continuous (body mass of an insect)? Only if you dichotomized that response so that it becomes binary (e.g., 0 = body mass less than some cutoff; 1 = body mass greater than or equal to that cutoff) and analyzed it with a generalized linear mixed effects model would you want to exponentiate the model coefficients and interpret them as odds ratios. (Dichotomizing a continuous response is bad practice anyway.) – Isabella Ghement Jul 20 '19 at 18:57
• If your group variable is coded as a factor using dummy coding, your reported effects will be represented as the difference in mean body mass values between the non-reference treatments and the reference treatment (adjusted for the random block effects). With dummy coding, one treatment is set aside as a reference and the remaining treatments are compared against it. – Isabella Ghement Jul 20 '19 at 19:00