Hi
I have a dataset in SPSS where a couple of variables has very few answers. There is not missing data its just that even though the variable could contain any number above 0, which would make the variable continuous, many participants actually answered zero. One such variable has only the numbers 0, 1, 2, 3, 4. Even tough it was possible to answer any number above zero would this make the variable ordinal instead of continuous? Should it be treated as such?
On a related note, on one variable only 7 participants provided an answer above 0, everyone else (N = 136) answered 0. How does this affect the analysis of variance? To me it seems like there is not enough variance to analyze it properly.
Any thoughts on this? Articles on the subject would be great.
Thanks
Hi! Have you thougth about perform a zero inflated regression such as zero inflated Poisson or zero inflated negative binomial?
Hello Daniel Pettersson. What are the variables? Are they counts, as Ana Carolina Martins guessed? Or are they something else? (By the way, note that counts are discrete, not continuous.)
You've told us (in a roundabout way) that n = 143. Can you tell us about the explanatory variable(s) too? And how many outcome variables do you have? (I'm thinking about the multiplicity problem here.) Thanks for clarifying.
Ana Carolina Martins Thanks, I have not heard of these things before which might make it a bit tricky, but I will look into it. Is there some literature you could recommend?
Bruce Weaver The independent variable which gets a lot of cases with zeros was coded interview data. This specific variable was number of times the participant said I don´t know (IDK) in response to a question. The result was that out of 136 participants, about half (if I remember correctly) made no such response. The few cases with the most IDK answers only made 4 such responses. So the variable it self, although it technically was possible to give plenty of IDK answers, actually only has four levels now. Using this as a dependent variable in an ANOVA just felt odd to me. It seems to me that this IDK variable is more like an ordinal variable, or perhaps a very limited interval variable since the numbers are actually representing amounts, meaning that 2 would be twice as many as 1. Still, the range is what I question here.
The other variable in question was also coded interview data to be used as a dependent variable. Participants answers were coded for subjectivity (i.e., number of details reported which would be considered subjective and therefore unverifiable). As it turned out most people reported no (zero) subjective details. Out of 136 participants only 7 reported subjective details, leaving the vast majority of cases at zero on this variable. So here I also thought what the heck is the point of adding this as a dependent variable in the model, how could such a non-informative variable provide any insights? Even if all 7 cases belonged to the same condition I would not feel confident stating that that condition resulted in more subjective details reported. There is so little data. But I´m not entirely sure I need to be so cautious which is why I ask.
I have several continuous (supposedly) DV´s, 2 categorical IV´s with three levels each and 1 continuous covariate. I was thinking about running either a MAN(C)OVA or a mediation analysis with AMOS. But out of all the DV´s some have the issue I mentioned above which makes me cautious, and hence Im reaching out to the fine people here on Research Gate.
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero