Logistic regression, but no negative responses?

[bernstml19]

So I am trying to model who is likely, from a list of addresses, to go to a store. I have lots of criteria about the people who have visited my store, but no data about who DIDN'T visit my store. I know if I had a 1,0 variable (visited store or not) that I could do logistic regression, but I have no idea even how many people might be out there for a given store.

Is there any way to come up with a certain response score that says, address with this distance from the store and this census block will be more likely to come to the store than another in a different block + distance?

Thanks,
Matt

 

[hlsmith]

Well there will be many issues such as who is eligible to come to your store. However if you knew who was eligible to come to your store and who did, plus the same characteristics about all of the people then you may have enough info to play around a bit.

 

[bernstml19]

Well there will be many issues such as who is eligible to come to your store. However if you knew who was eligible to come to your store and who did, plus the same characteristics about all of the people then you may have enough info to play around a bit.

So I have no idea who is "eligible". I know the count of how many people are in each census block - but I won't know their individual distance to the store. I do not have the negative responses.

 

[hlsmith]

Can you survey your clientele and make assumptions about them, say male age 20-40, live?, etc. Then look at census tract data. I am sure there are piles of resources for you some where, since business continuously target advertisement campaigns to folks based on some type of logic or strategy.

What is your end goal, to understand who comes in and make inferences/intimations about the others that don't come in?

 

[bernstml19]

I ultimately want to make a response score - odds that a person will come into the store based on their address and other factors. I only have a list of who came into the store, their census block and some other metadata - I don't have a list of who "didn't" come into the store.

 

[noetsi]

Basically you are trying to create a sampling frame - a list of the possible number of people who might come to your store. Since you know who does come to your store, if you know this you know who did not come to your store. There are two possible strategies that come to mind although one is probably only possible if you work for the government. That is to get a list of who lives in the region. Various organizations would have this, for example utilities, tax appraisers etc but you probably can not get access to this [again unless you work for the government doing a government project].

There are sampling technques that essentially create blocks and then randomly select geographic subunits within those blocks [for example cluster samples]. You then do polls within those sublocks [did you come /not come to the store although in this case you could just ask them their name]. That might be to costly for what you are doing.

In days of yore you could just get a phone book and do this I suspect, but so many people don't have land lines this won't work anymore. You might ask around to try to find some source in your area that does get at this type of information.