0
$\begingroup$

I have done an experiment with an independent and dependent variable. I repeated the experiment 5 times. Obviously all the repeats will be different from one another, some of them significantly so. However it is the pattern which I am interested in and therefore want to control for the repeats. I was going to do an ANCOVA but the covariate isn't continuous. What shall I do?

$\endgroup$
1
  • $\begingroup$ I am using SPSS. Any help with how I could do this there would help. Also if my covariate was not significant do I remove it from the model completely? $\endgroup$ Apr 15, 2015 at 14:27

1 Answer 1

1
$\begingroup$

Under the assumption that you sampled your subjects from the same population (e.g. you repeated your experiments in the same subject pool), you could treat your data as one experiment. You could even include in your model a replication factor that tests whether there were differences between your replications. The results of the test on your independent variable would be saying that on average across all your studies was the effect of the IV significant.

That's the easy way, but probably not best since it is hard to say you randomly sampled from the same population (and that there was no one in more than one of your studies). The hard but better way is to conduct a meta-analysis of your experiments. In short, you'll get the effect size of your IV and its standard error from each experiment and analyze that as new data (e.g. with an n of 5 is the average effect size different from 0). There are a lot of other considerations that make it different from regular ANOVA/General Linear Models. This provides a good tutorial. There are plenty more on Google

$\endgroup$
3
  • $\begingroup$ I'm not sure what you mean by "variation may be lost." Meta-analysis takes variation into account (i.e. the standard error). If you mean variation in effects, that is exactly what it is testing. The big picture problem is there has to be reason for all the variation. It could be due to sampling, in which case you want to know the average effect. It could be because something changed between experiments. If you figure out what that is, you could include that in your model, which would make your results more rich. $\endgroup$
    – le_andrew
    Apr 15, 2015 at 15:08
  • $\begingroup$ Apologies, I pressed enter before I had finished. That makes sense, thank you. $\endgroup$ Apr 15, 2015 at 15:20
  • $\begingroup$ No apologies necessary. Now go have fun with your models. $\endgroup$
    – le_andrew
    Apr 15, 2015 at 15:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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