No announcement yet.
  • Filter
  • Time
  • Show
Clear All
new posts

  • Using variable measured at different levels in a regression

    Dear statalist,

    I am studying the effect of banking competition at the local level on the relationship orientation of banks.

    I have been asked to use the variable "relationship banks" as a dependent variable.

    Relationship banks are equal to 1 if the bank rated relationship lending as very important while lending to SMEs and zero otherwise. The data for this variable was obtained by interviewing senior credit officers or ceo's at the headquarter level.

    Therefore I am not sure how local banking competition could effect relationship orientation at the head quarter level. This is why I didn't use this variable in the first place. I was wondering if my logic is correct.

    I would be grateful for your thoughts on this .

  • #2
    Hello naveed,

    The variable of interest is measured at the corporate level rather than at individual bank branches? At what level are your predictors measured? If they are at the bank branch level, then you will either have to create means of each of the predictors at the corporate level to predict the corporate level outcome or you will have to use gsem or the external program Mplus to estimate a multilevel structural equation model in which outcomes can be at the higher level of nesting. Predictors can be at either level.


    • #3
      Hi thanks for your reply, the dependent variable is from survey of bank ceo's, banking competition is measured at the district level and rest of the control variables are measure at the firm level. I am confused as to why lets say firm level variable or banking competition can effect CEO perception of whether their bank is a relationship or a non relationship bank.


      • #4
        Maybe there is no theoretical reason that your firm-level variables should be related to whether the CEO believes their bank is relationship-based or not. I'm not familiar with your field and so don't know what the difference between a firm and a bank. It would be more helpful to me if you clearly laid out the multilevel structure of your data (including examples of variables measured at each level).


        • #5
          Once again this discussion has been really useful and thank you for your spending time on this. Let me try and explain this a little clearly:

          Relationship bank (binary variable)=1 if CEO claims that his bank is relationship based. This data comes from a survey of CEOs at the banks headquarter level.

          Bank Branch competition index is measured at the district level where the firm is located .

          Firm characteristics such as age, location, accounting variable (profit, leverage etc) comes from a firm-level survey across districts

          So essentially I have been approximately asked to estimate the following:

          relationship bank= a+b1Bank Branch Competition +b2 Firm Characterestics +u

          I am not sure why firm characteristics such as accounting variable or local banking competition should be related to relationship bank which is measured at a higher level.



          • #6
            Even if you used something like gsem or Mplus, a firm level variable, which I believe is at the lowest level of your data hierarchy, cannot directly influence a higher level variable (measured at the district or corporate level). The district-level mean of a firm variable can influence a district-level outcome and likewise, a corporate-level mean of a firm or district variable could influence a corporate outcome.

            For example, consider the firm-level variable age. At the corporate-level, you could calculate each corporation's mean age of its firms. You might then be interested in whether corporations with older firms, on average, are more likely to be relationship focused.


            • #7
              This has been very useful, you have confirmed a suspicion that I had about this problem. Thanks