# SPSS: transforming variables?

#### Jazz3

##### New Member
I am doing a single linear regression and multiple regression, but I have a question about my independent variable.

My independent variables (X) are between 0 and 100.
Some of my hypothesizes is that "A high X will lead to a lower Y"

High X = value above fifty to hundred
and Low x = value from 0 till 49

I assume I have to change my X in SPSS so it corresponds to high and low?
How do I do this? Or am I completely wrong? #### ondansetron

##### TS Contributor
I am doing a single linear regression and multiple regression, but I have a question about my independent variable.

My independent variables (X) are between 0 and 100.
Some of my hypothesizes is that "A high X will lead to a lower Y"

High X = value above fifty to hundred
and Low x = value from 0 till 49

I assume I have to change my X in SPSS so it corresponds to high and low?
How do I do this? Or am I completely wrong? You don't necessarily need to change it. I probably wouldn't. Here's why: 1) how did you determine that 50 or higher was "high"? If this is an arbitrary cutoff assigned without evidence to support the cutoff, this isn't really a good idea (you'll unnecessarily lose information and changing the cutoff can give different results). 2) If you leave X as is, 0-100, you can just test that its coefficient is less than 0, implying that as X increases, Y decreases, on average (which would answer your question).

Another thing to consider is the nature of the relationship.

1)Do you think that above some threshold (like 50), Y declines with a "jump"-- i.e. a discrete change (think of a stair case with only 1 step for this example)? This would imply that there isn't much change for values within the same group, but between the groups there is a noticeable difference.
-OR-
2) Do you think that the change in Y is more gradual when you change X? In other words, even though 50, 55, and 60 might all fall into what you would consider a "high" group, there is a noticeable and gradual decline in Y as you increase from 50,51,52...55...60... This would also more fit with the idea that going from 49 to 50 is not some discrete jump.

If number 1 is the case, then sure, make categories out of the X variable, but recognize the loss of information. Be sure that your cutoff can be justified with subject matter knowledge/theory rather than "well, I thought this would be a good cutoff."

If number 2 is the case, just use X as it is, without transformation. Do a one-tailed test for beta1 < 0 to test if Y decreases as X increases.

Last edited: