Hi,
I am working on my master thesis, I want to test if using different kind of messages results in a higher likelihood to accept online cookies.
To do this I have a model with two dependent variables and three moderating variables. The first dependent variable consist of three different messages: Message 1(control), Message 2 and Message 3. Participants only read one of the three messages in the online survey. The second dependent variable is a Likert scale based variable and is also a moderator. The other two moderators and the dependent variable are also Likert scale based. I already collected the data and now I want to analyse it, I was thinking of using an regression model, but my dependent variable is bimodal, in other words, my respondents likelihood to accept cookies were mostly very high or very low.
IV1: Three different treatments(dummy)(N=110 for every message)
IV2: Likert scale based(3 items) (α =.885)
M1: Likert scale based( 4 items) (α =.794)
M2: Likert scale based (3 items) (α =.888)
M3: Likert scale based (3 items) (α =.885) (same as IV2)
DV: Likert scale based (3 items) (α=.930) (bimodal)
I now the basics of statistics which I learned in my courses but I never had any lectures about bimodal distributed variables.
So my qeustion is: What are appriote ways to test my hypothesis with a bimodal DV?
If you need more details please let me know.
Thanks!
Daan
@DavidMorse, it sounds like the dependent variable is bimodal, not binomial.
@DaanOuwehand, just to be clear, if youare thinking about using ordinary least squares regression (OLS; the common multiple regression), the assumption is that the dependent variable is conditionally normal, not that the dependent variable itself is normally distributed. To assess this, usually we look at the distribution of the residuals of the analysis.
Imagine you were conducting a one-way analysis of variance with two groups that had a strong effect on the dependent variable. You might expect a bimodal distribution, maybe like this:
http://rcompanion.org/handbook/images/image095.png
In this case, a general linear model (e.g. OLS anova or regression model) would be perfectly legitimate, because the dependent variable is normally distributed conditional on the effect of the independent variable. The residuals of the analysis look like this:
http://rcompanion.org/handbook/images/image096.png
Hello Daan,
With a dichotomous outcome variable, (binomial) logistic regression would appear to be a viable option for analysis. Potential moderators can be investigated by including interaction terms. Here's a resource that gives suggestions for how to analyze and report your findings:
Article An Introduction to Logistic Regression Analysis and Reporting
Good luck with your work.
the bimodal problem is a mixture of sets for example response=p*X1+(1-p)*X2 if the DV has two modals just apply a mixture of two likert variables. the hard part is find p but u can just using bootstrap sampling and find them mean ...well its soluble..
@DavidMorse, it sounds like the dependent variable is bimodal, not binomial.
@DaanOuwehand, just to be clear, if youare thinking about using ordinary least squares regression (OLS; the common multiple regression), the assumption is that the dependent variable is conditionally normal, not that the dependent variable itself is normally distributed. To assess this, usually we look at the distribution of the residuals of the analysis.
Imagine you were conducting a one-way analysis of variance with two groups that had a strong effect on the dependent variable. You might expect a bimodal distribution, maybe like this:
http://rcompanion.org/handbook/images/image095.png
In this case, a general linear model (e.g. OLS anova or regression model) would be perfectly legitimate, because the dependent variable is normally distributed conditional on the effect of the independent variable. The residuals of the analysis look like this:
http://rcompanion.org/handbook/images/image096.png
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning