0
$\begingroup$

I am from computer science background and a bit new to statistics. Currently I am working for a online property portal and part of my work is to create indices for different statistical measures.

Demand Index:

Based on the number of views for a property and number of enquiries we would like to create a demand index. So we have the view and enquiries count for every day. What would be the best way for computing demand index? How do we assign weights to views count and enquiries count?

Also I think the Case-shiller index method is not well suited in this scenario and we have data for large number of propertied on a daily basis.

Improve this question
$\endgroup$

2 Answers 2

Reset to default
2
$\begingroup$

For each day, rank the number of views and the number of enquiries for the properties. Scale between zero and 1 by simple division. Add.

This gives you a composite index between 0 and 2. "0" if a property had the least views and the least enquiries, and "2" if a property had the most views and the most enquiries.

A property can score "1" if: it has the most views and the least enquiries; or the least views and the most enquiries; or the median views and the median enquiries.

This is just a simple unweighted sum of ranks.

I don't think a statistician can tell you how to weight the two measures you have. That's your job - maybe enquiries are more important to you than views? Then scale the enquiries rank. How much? I don't know - its not a statistical decision, its an operational one based on your belief in the significance of the measure.

If the two measures are perfectly correlated anyway, you don't need to do any of this. Plot a scatterplot, do the correlation, see if it is worth the complexity...

Improve this answer
$\endgroup$
0
$\begingroup$

I think you should try a regression approach. As I can see you have one response variable and two regressors, so the linear model could be a good approach. You don't need weights for each variable, the regression coefficients tell you how much every regressor influences the response variable.

Beware that, besides other hypothesis, the response variable must follow a normal distribution, so you should check for that in advance. If that's not the case and you want to follow the regression path, you have to go for a generalized linear model, but that is more tough since you are new to the science.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Hey, thanks @arpayon. I do not have the training data in the form of z=ax+by. If I did, I could regress and find suitable param values for a and b. But all I have is x and y and I am to derive an index representing z. $\endgroup$
    – Krishnaa
    Commented Sep 8, 2015 at 4:23
  • $\begingroup$ Ok, my fault, I thought you had also a z column. In this case a regression doesn't work because you need a response variable. I am sorry but I don't know how to proceed further :\ $\endgroup$
    – Arpayon
    Commented Sep 9, 2015 at 10:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.