Linear regression analysis - Urgent assistance required

bulu

New Member
#1
Hi

I am trying to examine the relation between ad. liking and attitude towards the brand. My hypothesis
H1:
"Commercials that are liked will lead to higher level of more positive attitude for the advertised brand"


Independent variable
Q- How did you feel about this commercial?
Answer OptionsResponse PercentResponse Count
Liked very much40.0%
Somewhat liked it37.8%
Neither liked nor disliked it13.3%
Did not liked it very much8.9%
Completely disliked it0.0%


Dependent variable
Q - How did you feel about the brand after having seen the commercial?
Answer OptionsResponse PercentResponse Count
Liked very much20.0%
Somewhat liked it37.8%
Neither liked nor disliked it40.0%
Did not liked it very much2.2%
Completely disliked it0.0%

Would linear regression analysis suit here? If so, I am bit lost how to do it?

Thank you for your assistance
 

terzi

TS Contributor
#2
Hi,

I don't seem to fully understand your study. At first, I' don't know exactly how you measured your variables, but since I consider them to be both categorical, I wouldn't recommend performing a regression analysis. In fact, your hypothesis only states about a positive relationship, so using a non-parametric correlation measure would be enough. You should study a little bit about correlation measures.

Now, you could perform a regression, although we would be talking about a totally different model, an ordinal regression, which is even more difficult and, in my opinion, not necessary. Probably a deeper analysis where you could get even more information is a Correspondence Analysis. Still, for your initial hypothesis, the correlation analysis should be enough.
 

bulu

New Member
#3
Thanks Terzi:)

You are right; I should have given a little bit more information.

Nevertheless, my understanding is that - categorical data involves placing things in a limited number of categories. In organizational research 'things' are people and typical category include gender etc. In my case, the data is continuous since I am using a likert scale.

What I had learned from my research professor ( a year ago ) is that if your dependent variables ( in this case - purchase intent) are in likert scale, linear regression analysis should be used.

My intention here is to examine the relation between advertisement likeability and purchase intent. Ad.likeability is the dependent variable (y) and purchase intent is the independent variable (x). If my data were categorical and I was examining relation between these two variables y = aX + b ( linear regression) may be enough. However What I have not mentioned earlier that there is another independent variable ( purchase frequency) which may have a moderating effect on this relationship. I therefore thought a basic linear regression should be done. I did so, using PASW 18, now struggling to interpret the result.

Appreciate your thoughts.
 

terzi

TS Contributor
#4
Oops, I guess you should seriously speak with your professor:). I'm not certain about the number of categories you used for measuring, but, in my opinion, you should avoid using Likert scale as if it was measured as a continuous response. Specially for modeling, since it is likely that the required assumptions won't be filled.

Check this link for more information:

Code:
http://www.analysisfactor.com/articles/Likert-Scale-Data.html
If you are using a linear regression with your variables be extremely careful.
 

CB

Super Moderator
#5
Nevertheless, my understanding is that - categorical data involves placing things in a limited number of categories. In organizational research 'things' are people and typical category include gender etc. In my case, the data is continuous since I am using a likert scale.

What I had learned from my research professor ( a year ago ) is that if your dependent variables ( in this case - purchase intent) are in likert scale, linear regression analysis should be used.
Likert scale data is not continuous. It is ordinal data, which in some cases you may be able to justify *treating* as continuous/interval. That sounds like some dodgy advice from your prof - linear regression can be suitable for use with Likert scale data, but it depends a lot on how many points are on the scale, how the response points are labelled, and the distribution of responses on the scale.

In this case, with 5 response points you may be able to justify treating the data as continuous - but the skewed distribution of IV data points is probably going to result in violation of the normally distributed errors assumption for OLS regression. Ordinal regression could be an option but you need to be able to meet the test of parallel lines assumption.

Non-parametric rank based correlation (e.g. Spearman) does seem like a good place to start to see if there is a bivariate relationship between the variables at least :)
 

bulu

New Member
#6
Likert scale data is not continuous. It is ordinal data, which in some cases you may be able to justify *treating* as continuous/interval. That sounds like some dodgy advice from your prof - linear regression can be suitable for use with Likert scale data, but it depends a lot on how many points are on the scale, how the response points are labelled, and the distribution of responses on the scale.

In this case, with 5 response points you may be able to justify treating the data as continuous - but the skewed distribution of IV data points is probably going to result in violation of the normally distributed errors assumption for OLS regression. Ordinal regression could be an option but you need to be able to meet the test of parallel lines assumption.

Non-parametric rank based correlation (e.g. Spearman) does seem like a good place to start to see if there is a bivariate relationship between the variables at least :)
Thank you so much.

You advise is vary valuable. Also, may be I have misunderstood or have a vague memory of what my prof. said, but that was then.

It seems that I will have to bin two days work and start from zero. Thank you once again.