Proc Corr with Character Variables

[RamKumar]

I do services for banking and I need help to find the correlation between two variables. Interestingly I need to do with character variables and below are the sample data.

Our task is to sell loan products. Eg: many occasions people who brought mortgage loan is likely to buy auto loan.

Similarly I need to identify the loan products which has correlation.

product_dsc    loan_amt

AUTO LOAN       26,698.65

AUTO LOAN       15,000.00

AUTO LOAN     15,200.00

Used Auto        7,194.00

Personal loan   1,000.00

Used Auto     23,000.00

Used Auto    12,400.00

Any help would be really appreciated.

[Ksharp]

You need to focus the count number of person who have a loan  not load amount , then use proc freq or proc corresp  .

[RamKumar]

Thanks Ksharp.

So you're suggesting me to use proc corresp instead of Proc corr because of character variables? My assignment is on correlation and I will be worried if we don't have relationship between proc corresp and proc corr

[Ksharp]

Yes. proc freq also can compute person,spearman correlation coefficient for category variable ,but you need frequency of people , check documentation.

BTW, you also could try proc corr ,but that will give you person product-moment correlation coefficient  if you have data like

name   home_load auto_load

Arthur.T   23    45

Peter.E   34  45

KSharp   56 78

Xia Keshan

[stat_sas]

Hi,

Change your data structure to run correlation analysis. Identify people who are having more than one loan (auto_loan, used_auto etc.) then using proc corr can give some idea about the customers' behavior.

Id      AUTO_LOAN       Used_Auto     

1       26,698.65            7,194.00

2      15,000.00            23,000.00

3      15,200.00            12,400.00