# Quadratic Regression with a proportion scale in SPSS?

I want to perform a Quadratic regression in SPSS. The dataset includes two variables:

• Independent variable = the proportion of hours worked from home -> (ranges from: 0.00 - 1.00)
• Dependent variable = work engagement (ranges from: 0 - 5)

My problem is how to perform a Quadratic regression in SPSS?

The problem: one of the steps of a Quadratic regression analyses in SPSS is to compute a to create a variable that is squared. But, with a proportion-scale (0.00 - 1.00), this is very different as the size decrease. For example 0.2 - 0.04.

Question:

• What is the best way to perform a quadratic regression analyses with a proportion data scale as independent variable?
• How to compute the new squared variable?

Example of the normal procedure (not applicaple here, because square is not possible): https://www.statology.org/quadratic-regression-spss/

• "But, with a proportion-scale (0.00 - 1.00), this is very different as the size decrease". Why this is a problem?
– mkt
Jun 24 at 4:17
• Well, the size of the proprotion decreases when squeared: For example: 0.80*0.80 = 0.64 0.40*0.40 = 0.14 0.20*0.20 = 0.04 The new computed (squared) variables are totally different and decreases in stead of multiply with squared function. See attached the example, 5*5 = 25 - 6*6 = 36 -> it increases. Jun 24 at 7:20
• If you wanted to, you could rescale 0-1.00 by multiplying by 100 (so the range would become 0-100), and then the squared value would not decrease any more. But I still don't get why this decrease is a concern.
– mkt
Jun 24 at 9:28
• What is the function to compute a new variable to square when the proportion value decreases? Jun 24 at 13:54
• The difference in sizes between proportions and their squares is of no consequence (it's actually meaningless, since the two are in different units anyway). What really matters is that you should be concerned about any sizable variations in the duration of time in which each proportion was measured, because that affects the relative precision of the data and ought to be accommodated by adopting a suitable model or at least with weights in an ordinary least squares regression.
– whuber
Jun 24 at 14:24