Thread: Multiipe regression equation query

Suppose I'm looking at frequency of domestic garden watering and through running a t-test have determined that there are two periods in time which differ significantly in garden watering frequency - spring/summer and autumn/winter. I've also run correlations and determined that the key predictor variables in both cases are temperature and socio-economic status of the household.

Now if I want to perform a multivariate regression to predict frequency of garden watering, do I need to create two regression equations, one for spring/summer and one for autumn/winter given that the t-test showed the distributions are different, or do I just create one regression equation for the whole year?

Thanks
-Rob

If i remember correctly you can either fit 2 seperate lines or a combined model as follows:

y = frequency
X1 = temperature
X2 = socio-economic class
X3 = Season indicator (0=spring/summer, 1=autumn/winter)



This will give different slopes and intercepts for each value of the indicator variable. I dont remember, but i suspect the two approaches give the same answer unless you can use some process to reduce the number of terms in the combined model.

Thanks (again) Spring. You're fast becoming my MHF Hero!