1
$\begingroup$

Question: Did plants experiencing treatment A vs. treatment B receive significantly more or less observations with zero visits from pollinators?

Methods: Plants were clustered by similar physical traits (2 plants/cluster). One plant in each cluster received treat. A, while the other received treat. B. Each individual plant was then observed 14 times over the period of one week with video cameras.

Preferred platform: R

Looks like McNemar's test only works if each subject is only measured 2 times, while in my study each individual is measured 14 times. Any suggestions at all would be appreciated!

$\endgroup$
2
$\begingroup$

Cochran's Q test (1950) is akin to a repeated measures ANOVA for binary outcomes. With only 2 repeated measures, Cochran's Q gives the same results as McNemar's test.

My answer to a question about the Effect size of Cochran's Q details effect size calculations for Cochran's Q. I have implemented the test in the cochranq package for Stata (within Stata type net describe cochranq, from(http://alexisdinno.com/stata)).

This all said, it is not clear to me that you do not actually need a regression model, since you apparently want to account for a treatment variable, as well as a measurement number (unless specific treatment goes along with specific time of measurement). If this is the case, you may want to explore using a multilevel logistic regression model, where time of observation is nested within plant clusters, and treatment is a plant cluster-level independent variable.

Cochran, W. G. 1950. The comparison of percentages. Biometrika, 37: 256–266.

$\endgroup$
2
  • $\begingroup$ Thank you for the response, to answer your question- I think you're right in looking for a multilevel model- I am not really interested in change over time, just overall response to treatment, though I certainly do need to look it over- repeated measures was used only to get an adequate sampling from each tree, due to limitations of the system (the treatment process was too time-intensive to take only one measurement per individual, and there were limited individuals with which to work). Thus, I think that perhaps the regression model approach is the way to go. $\endgroup$ – kehau Mar 19 '15 at 5:53
  • $\begingroup$ @kehau Note that just because a multilevel model incorporate the repeated measures structure, this does not mean that you must necessarily include time itself in the model (i.e. you compelled to obtain fixed effects estimates for time simply because period of observation is part of your data hierarchy). Also: If you like my answer, do feel free to up-vote by clicking on the up arrow and or accept my answer by clicking on the check-mark at the top left. $\endgroup$ – Alexis Mar 19 '15 at 16:32

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.