I need to perform a binary regression analysis to determine the effectiveness of different doses of a certain drug on sex ratio (male or female) of a fish. Can anybody assist me how i can arrange the data (dosages, male, & female) in SPSS for analysis, and which important information do i need to present to indicate the dosage responsible with masculinization or feminization?
Hello,
Typically when you have a dichotomous categorical variable, such as the sex of fish, and want to run a regression, regardless of the type, you would want to have a column of data representing your outcome measure, a column for your continuous dose variable, and then a "dummy" variable that either has a value of "0" or "1". This dummy variable indicates, in the context of your problem, the sex of the fish and you can choose whether "0" represents male or female with "1" representing the other sex.
Usually, the sex assigned to "0" is considered the baseline, but this doesn't have to be. Say you were running a multiple linear regression where you assigned males to "0" and females to "1" in the dummy sex column and obtained the following equation (the beta coefficients and intercept are made up for the example):
y = 1.0 + 0.5(dose) + 2(sex)
Then, assuming higher values of y indicate greater effectiveness of the drug, holding the dose of the drug constant at a value of 10 units would indicate an estimated response of 6 effectiveness units if the fish was male (sex == 0) and an expected response of 8 if the fish was female (sex == 1). If you had reversed the "dummy" assignment for sex, then the beta coefficients would be different, but should give you equivalent expected values.
Here are two links for further information; one is a series of videos talking about dummy variables (with more than two categories) in multiple regression and the second is SPSS's tutorial on the subject.
Hope this helped.
http://www.spss-tutorials.com/creating-dummy-variables-in-spss/
https://www.youtube.com/watch?v=R0qc4rzr9ik
What Jeremiah explains is about binary logistic regression. I do not think Naftal is asking about this, because his response variable is a percentage (sex ratio) not a binary one.
According to what I understand from Naftal's question, a binomial logistic regression is needed. I do not have SPSS access now and I do not remember exact setup. I attached a video which should help you.
So what you need to do is a binomial (not binary) regression (logistic or probit). Probit is generally done for dose-response relationships. You could do logistic binomial regression with using Generalized Linear Models menu in SPSS which is more complicated.So go with the probit.
As I remember, you should enter total number of fish together with the number of males as numbers not fractions and SPSS will compute necessary percentage.
Yes, I find another source now which explains in detail
https://www.youtube.com/watch?v=TxwvDgoPKHE
http://userwww.sfsu.edu/efc/classes/biol710/probit/ProbitAnalysis.pdf
Is your dependent variable a proportion? e.g. % of females
If so, it sounds like it might be difficult to justify this as a binomial distribution. Some clarification on the nature of your dependent variable would help identify appropriate methods of analysis.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS