I am trying to determine the best regression model to use to determine if there is a relationship between clicks (independent variable) (represented as a whole number) and sql % (sql/sal)(represented as fraction)(dependent variable).

Can any explanation of this data be valid if I were to run a linear regression on these two metrics (clicks and sql %)?

As I pointed out, sql % is the output of dividing sql/sal. Both sql and sal are represented in my table. Would using either metric in the regression act as a substitute for the dependent variable in the original regression I'm trying to run (clicks,sql %)?

I currently ran a linear regression on clicks and sql % and there is a R-Square of 0.018 which makes sense when looking at the plotted data points and clearly seeing that there is no linear fit.


date    sals    sql %   sqls    clicks
9/1/17  31      0.32    10      18
9/2/17  9       0.11    1       17
9/3/17  10      0       0       11
9/4/17  15      0       0       11
9/5/17  41      0.41    17      7

Since your variables $sqls$ and $sals$ both seem to be counts you could choose Poisson regression. The disadvantage of linear modelling of the $sql %$ is that it is bounded at zero and unity whereas linear regression gives predictions potentially outside that range. If you use Poisson regression you model $\log(sqls)$ as a function of $clicks$ using $\log(sals)$ as an offset. This gives the multiplicative effect of $clicks$ on $sqls$ by exponentiating the coefficient. There are many posts on this site going into more detail about Poisson regression. It is widely available in statistical software like Stata or R.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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