If your hypotheses are that each is associated with the behavior, then this is four correlations. If your hypothesis is that the four together predict the behavior, then probably include their interactions and do a multiple regression. If your hypotheses are that each of these, after conditioning on the other three (but not others) is associated with the behavior, that is an odd set of hypotheses, but you'd look at the individual coefficients from the multiple regression. From the way you have phrased this I would guess 4 correlations and then adjust for multiple hypotheses.

multiple regression would be my choice here, because I would want to statistically adjust for the other IVs

Daniel Wright, so 4 correlations would be okay. what do you mean with adjust for multiple hypothesis?

If you have four hypotheses/p-values in the same "family", and you want to set the familywise alpha at .05, you would need to adjust the alpha for the individual correlations. If you think that is appropriate though. There is a trade-off when adjusting to keep that overall Type 1 error rate at .05 and the Type 2 rate. Anyway, a common approach with 4 p-values would be Holm's procedure. Say your four p-values are: .001, .01, .02, and .04, you could stick these into the R function p.adjust and get the results:

> p.adjust(c(.001,.01,.02,.04))

[1] 0.004 0.030 0.040 0.040

Note that in some older texts Bonferroni's is suggested. This yields:

> p.adjust(c(.001,.01,.02,.04),method="bonf")

[1] 0.004 0.040 0.080 0.160

>

so you'd have two fewer "significant" findings if that aspect is important to you (but see all the hoopla in Nature last week on p-values).

I agree Dr. Daniel Wright

Hi Kim,

Perhaps legacy methods will produce results which you find satisfying. However, historical methods such as correlation (more than a century old) and multiple regression generally fail to accurately predict scores which deviate from the sample mean. Predictive accuracy is improved using a modern machine-learning paradigm:

https://odajournal.com/2013/09/20/maximizing-the-accuracy-of-multiple-regression-models-using-unioda-regression-away-from-the-mean

https://odajournal.com/2016/09/19/novometric-analysis-with-ordered-class-variables-the-optimal-alternative-to-linear-regression-analysis

https://odajournal.com/2017/04/18/what-is-optimal-data-analysis

In your question you used the term “predict”, if you intend to imply that each of the four variables (above mentioned) predict a certain “behaviour variable” then in my opinion, multiple regression will be the appropriate method. Because regression involves prediction... estimating the impact of indepedent variables on dependent variable. But if your research question/hypothesis aims to analyze the association or co-dependency between these four variables and behaviour then you can rely on correlation only. It totally depends on your hypothesis.

Crisis and emergency alert http://youtu.be/Ng1-KJueYiU Time for the people to stand together to bypass, help us build the bypass. We have the foundation's know