# Logistic regression with three possible values for the dependent variable

#### anders

##### New Member
I have a set of paired X and Y values, where the X values vary from about 6 to about 100, and about half of the Y values are zero, and the rest are either positive or negative values. The absolute values of Y vary so much that a normal linear regression is useless. It very clearly looks like, however, that with increasing X leads there is an increasing probability that Y is positive, and this is what I want to prove. I don't know how to calculate the statistical significance of this, however. If there were no negative values, I could use a logistic regression, but in this case it is important to take into account also the negative values. See graph below. How can I go about in order to do a statistical test for this? #### fed2

##### Active Member
why must the negative values be taken into account? can't they be grouped with the 0's?

#### obh

##### Well-Known Member
Hi Anders

The absolute values of Y vary so much that a normal linear regression is useless
When you transform to -1,0,1 you lose some data.
Did you try any other transformation toward the linear? for example sign*log(abs(x))

If there were no negative values, I could use a logistic regression,
Logistic regression is for non-ordinal categorical DV.
For ordinal DV you should use the ordinal regression

But I assume a better transformation may be better ...

#### anders

##### New Member
Hi Anders

When you transform to -1,0,1 you lose some data.
Did you try any other transformation toward the linear? for example sign*log(abs(x))

Logistic regression is for non-ordinal categorical DV.
For ordinal DV you should use the ordinal regression

But I assume a better transformation may be better ...
Hi, thanks for the suggestions. I tried an ordinal regression on SPSS, but I don´t understand the output at all.....

#### hlsmith

##### Less is more. Stay pure. Stay poor.
If the 1,0,-1 values are truly the DV and have an ordered interpretation, ordered multinomial logistic regression could be used. However, does the above image display all of your data? If so, a group with only three values may be highly suspect to making inferences or having confidence in estimates.

#### Karabiner

##### TS Contributor
. How can I go about in order to do a statistical test for this?
In my opinion, we need more knowledge of what the data mean,
otherwise we are referred to guessing. So hat is your research
question? What are these X and Y variables, and how were they
measured?