1
$\begingroup$

I am a biology major, and will be running a study and am unsure of what statistical test to run. Here is a summary of what we will be doing:

We are testing to see if there is a correlation between weight gain and amount of light in crickets. Using 20 crickets, they will be split evenly into two groups: a control group getting normal daylight hours, and an experimental group that will be in total darkness. The project will go on for 2 weeks, and each cricket will be weighed once a day for the 14 days. I imagine the variables can be plugged in as idno (1-20), light (1=control, 2=dark), and day (1-14, each weight will be recorded here).

So, which test should I run? I thought possibly a repeated measures ANOVA but after running, that, it doesn't seem to be testing for what we're looking for. To state it again, we are looking for a significant difference in weight gain between crickets kept in normal lighting conditions and those kept in 24/7 darkness.

$\endgroup$
1
  • 1
    $\begingroup$ How are you weighing them in the dark? You wouldn't be able to read the instruments. Why doesn't a repeated measures anova deal with the question of interest? What is it giving you? $\endgroup$
    – Glen_b
    Mar 24, 2015 at 22:29

3 Answers 3

1
$\begingroup$

Since the 2 groups are not paired, paired t-test would be inappropriate. Since the number is relatively small, a non-parametric test would be better. Since weight gain is the issue, final weight gain (at 14 days) can be taken ignoring values taken during the study.

Mann-Whitney U test for unpaired data could be applied here for 'weight gain at 14 days' in dark vs light groups. http://en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U_test

$\endgroup$
0
$\begingroup$

I'd agree with your gut thought of a 2 factor repeated measures ANOVA, where you are interested in the outcome of the treatment/control effect or, depending on your research question, the interaction effect between time and treatment/control. Such a design may have limitations, though, especially with your proposed sample size... Another thought could be a t-test with "weight gained" as the dependent variable and treatment/control as the independent variable.

$\endgroup$
1
  • $\begingroup$ Data on "amount of light" does not seem to be available. $\endgroup$
    – rnso
    Mar 25, 2015 at 13:16
0
$\begingroup$

If you are considering using pre/post scores instead of all 14 data collection points (seems like it would still adequately answer your question), a discussion of possible methods for analysis occurs on the following thread: Best practice when analysing pre-post treatment-control designs.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.