See the attached paper by Shmueli for some very good advice. If you are interested in building a predictive model, I suggest using lasso variable selection as described in the other paper below. Best, David Booth

If your variables are measured in the same way (it seems to be so, and it should be so if you want to compare the effects), just merge/append the databases.

Obviously, include a dummy first in each of them: D: 0= city A, 1=city B.

Let us say your model is P = b0 + b1*Inc + b2*Int + b3*Nhou + b4*Dpop

Then, there are several alternatives:

a) Include interaction terms:

P = b0 + a0*D + b1*Inc +a1*D*inc + b2*Int + b3*Nhou + b4*Dpop,

where, remember, D = 1 if city=B and 0 if city=A.

-- a1 captures the change in the slope of the price-income estimated relationship in city B vis-à-vis city A.

-- Hence, this allows you to test whether the effect of income is the same in both cities (a1=0) or not (is a1>0, the effect of income is larger in city B; if a1

This book might help you

Structural econometric modelling: Methodology and tools with applications under EViews

http://www.eviews.com/StructModel/structmodel.html

Thanks everyone for your answers,

Andrés im currently including interaction terms. Since the only variable with the same data sets for both cities is interest rate, income population and new houses are data sets based on each city.

This is my equation where d1 is the dummy

priceindex popgrowth income d1 realrate supply c d1*income

Do I run each interactive term individually or all together to be able to compare each variables effect.

Thanks in advance

Didrik

I opened your dataset, I saw that all these variables are in both datasets. I included an interaction term only with one variable. Including several interaction terms in the same model may be tricky due to eventual collinearity problems (among the interaction terms), especially if you do not standardize the variables.

m1 - without interactions

m2 - interaction with x1

m3 - interaction with x2

...

Many articles following this strategy are published in top journals in pol sci, my discipline.

If u feel this is not ok for your research area, then the idea would be

a) check collinarity among interaction terms (say with VIF)

b) if it looks bad, standardize, and check again.

If u know how to check, do it. I cannot check it today. It's not too difficult, but I promised to send the data analyses of a paper before midnight... still 1h 15 to do it!

By the way. I see your data involves time. This complicates a bit the story, since regressing without acknowledging the time-structure of the data may be a problem.

And second additional comment. Your N is not too high.

Yes N is not high but I can't get it any higher so it is fine according to my professor. Do I really need to take time into consideration? Should I use Robust Standard errors since its panel data?

Thanks for your assistance hope your papers going good :)

time is complex. forget about it 4 z moment dealing w many issues at z same time is not a good idea. i just pointed 2 z N and 2 z time issues so that u remember 2 consider them later.