# Linear regression- If you ever wanted a chance to save somebody's life...this is it!

#### melsuett

##### New Member
Hi guys,

I will be forever grateful for some help with SPSS and linear regression.
Here is my data set:

-I have questionnaire data from employees and managers. Both filled in the same questionnaire. Sample size all in all : 32
-The questionnaire has 8 themes and demographic data. Each person has a score for each theme

I want to see if there are relationships between an employee's theme 1 and theme 3, for example. And then if there is a relationship between the employee's theme 1 and the managers' theme 1 scores.

How do I do that?

I started with doing a Pearson's correlation. There were several significant correlations, more than 20 i think. Now I was going to follow this up with a simple linear regression. Does that even sound okay?

Do I have to do one for each signifiant pair? I can't do a muliple regression because my sample is too small right?

And next question: In SPSS I have one colum for the position (either employee or manager) and labeled them 1= employee 2= manager. If I do correlations, how can I tell SPSS to correlate the 1s with the 2s in a specific theme??
The Pearson's corrleation I did- that never differentiated between managers and employees ..i suppose...? How do I tell it to only correlate the themes in the managers and then employees seperately?

#### Ilya Maclean

##### New Member
Hi guys,

I will be forever grateful for some help with SPSS and linear regression.
Here is my data set:

-I have questionnaire data from employees and managers. Both filled in the same questionnaire. Sample size all in all : 32
-The questionnaire has 8 themes and demographic data. Each person has a score for each theme

I want to see if there are relationships between an employee's theme 1 and theme 3, for example. And then if there is a relationship between the employee's theme 1 and the managers' theme 1 scores.

How do I do that?

I started with doing a Pearson's correlation. There were several significant correlations, more than 20 i think. Now I was going to follow this up with a simple linear regression. Does that even sound okay?
Not really for two reasons. Firstly, a regression gives the same P value as Pearson's correlation, it just has a different assumption - namely that one variable causes changes in the other. Consequently it gives you an equation that allows you to predict the value of Y from any given value of X rather than an r-squared value. Surely an employers score does actually depend on a managers score (or vis-versa)?

Secondly, because you're dealing with scores which are ordinal rather than continious you should be using a non-parametric test such as Spearman's rank instead of Pearson's.

Do I have to do one for each signifiant pair? I can't do a muliple regression because my sample is too small right?
See above for why probably don't want to do regression at all. Additionally, whether you would do any test between each pair is a question that can't be answered by statistics, but only by whether you are hypothesising that a relationship between pairs might exist. You should also be aware that, if you doing lots of pair-wise comparisons, the law of averages means that some will have P values less than 0.05 purely by chance. You probably need to apply a Bonferroni correction. Lastly whether or not you chose to do multiple regression as opposed to pairwise regressions does not depend on your sample size, but on what you're testing for. A multiple regression is used if you expect one variable to be affected by more than one other.

And next question: In SPSS I have one colum for the position (either employee or manager) and labeled them 1= employee 2= manager. If I do correlations, how can I tell SPSS to correlate the 1s with the 2s in a specific theme??

The Pearson's corrleation I did- that never differentiated between managers and employees ..i suppose...? How do I tell it to only correlate the themes in the managers and then employees seperately?
Have seperate columns for each and match them side by side.