Reference to justify the use of bootstrap

Dear all,

My colleagues and I conducted a research involving a dichotomous outcome (i.e., the detection of an unexpected event in a particular scenery). We assigned participants randomly in one out of three condition. Our hypothesis is that one of this condition will increase the detection of the unexpected event. We decided to analyse the data using bootstrap analysis to estimate detection rates and CIs (rather than binary logistic regression). One reviewer of our first draft asks for references to justify for the use of bootstrap in this situation. Could you, please, direct me to such references?

Thank you.


TS Contributor
You could argue that since CIs from binary logistic regression is valid only in the asymptotical case (not sure, so please correct me if I'm wrong), bootstrap is preferable in your specific case.
@Englund : Thank you for your answer.

@hlsmith : Thank you too. I am no expert in bootstrapping methods, so I did the most straightforward way I could think about. I regressed my DV (detection: 0=No, 1= Yes) on my 3 (Condition : Control, semantic priming, goal priming) x 2 (Task difficulty: Easy vs. Hard). The regression does provide regression coefficients related to detection rate, and the bootstrapping is used to determine the CI of these coefficients. I used a non-parametric bootstrap method (I guess) since I used 5000 or 10000 pseudo-samples (or re-samples? I don't know the terminology), and determined the (bias corrected) 95% CI using these samples.

Basically, I adapted the procedure Edwards & Lambert (2007) used for (moderated) mediation analyses to my case. My idea was, as bootstrapping is recommended in the case of mediation due to the violation of the normality assumption, and as the problem with dichotomous DV is exactly this violation, I assumed that bootstrap could be used in my case. Since the results were plausible and consistent with those obtained through logit, I decided to present the bootstrapping results in the paper, as I believe it is a powerful, underused, straightforward procedure. Maybe I am wrong. If yes, please say so, and please provide some references so I can understand what my mistake is.

Thank you. :)