I have the following variables:
1) External pressure from
a. Customers
b. Regulations
c. Competitors
d. Partners
e. No pressure
2) External pressure from
a. Employees
b. Investors
c. CEO
d. CSR
e. No pressure
3) Do you have specific goals for the following:
a. Resource consumption
b. Energy consumption
c. Water consumption
d. Alternate energy
e. No
I want to show the association between these three variables and then test the regression of question1, and 2 on question 3. The data for each option of the above questions is coded as 1= yes, 0 = no). Would you please mind how we can regress question 1 and question 2 on question 3?
First of all, you need to check for any existing relationships between the variables (it's obvious).
Therefore, it is necessary to calculate the vector of means, the variance-covariance matrix, and the correlation matrix.
Later, you can graph using scatterplots, which plot scatterplots for all pairs of variables in your matrix.
The rest will come by itself.
Thank you Gioacchino de Candia for your reply. I mean which regression we will use as I intends to jointly consider question 1 and 2 and then regress it on question3. I am interested to know when external pressure from stakeholders (question1) combines with internal pressure from stakeholders to affect question 3. How would be that if I cluster all the companies based on question 1 and question 2 and then associates those clusters with question 3? Would you please kindly help here whether it is justified to do so?
Fazli Haleem , Is it the case that for each of these questions, respondents can choose more than one answer ?
If so, the easiest approach is to consider each option as a separate binary variable. It's helpful to look at the correlation among each pair of these binary variables (the phi statistic for a 2 x 2 contingency table). These simple correlations may tell you what you actually want to know.
To conduct a larger regression, the easiest approach would be to use binary logistic regression, treating each option in Question 3 as a separate dependent variable. Be cautious about having multiple correlated independent variables in the model.
Please note that the second group of external pressure is not external, in principal it is internal pressure. Please look into this.
Thank you for your reply. Yes for each of the sub options such as a, b, c,d,e of question 1, 2, and 3 respondents can chose more than one answer.
One solution is to treat all the suboptions each an independent variable (binary) and then do interactions with the sub options from question 2 and then regress them on the sub options of question 3. However, I am more interested to investigate whether question 1 and question 2 is inline (together) and then see its impact on question 3. For that purpose if I do cluster analysis and identify clusters given the question 1 and question 2 and find the association of these clusters with the question 3. How do you see this solution?
I think clustering the respondents by their profiles on Questions 1 and 2 makes sense. But I don't know what kind of cluster analysis will work for what is effectively several binomial variables.
In any case, I wouldn't skip looking at all the bivariate correlations. This may tell you what you actually want to know.
We can cluster on any attributes and in our case clustering will help to give us clusters where there will be only external pressure (question1) from various stakeholders (a, b, c,d,e), only pressures from internal stakeholders (question2) (a, b, c, d, e), and combine both. Then we can regress these clusters on question 3. I think the binary variable should not be an issue each option is taking a value of 0, 1.
(this is assuming I understand that each respondents was presented with these as 15 Y/N questions).
First step. Check that the e option is used if and only if a-d are all "no." If there are cases (and how many, you didn't say) with this inconsistency, you should figure out (did you run think-alouds or some other method like that when pilotting?). You will need to decide what to do with these.
Second step. You are left with 8 binary variables in q1 and q2, and 4 in q3. Including interactions just of the q1/2 variations, that is about a thousand combinations for the four q3 variables. So, the fishing for large associations approach mentioned by some commentators may have some issues. It will be useful for you to constrain these by whatever hypotheses you are interested in. If you share this with us you might get more directed answers.
Thank you Daniel Wright for your reply. The main hypothesis is: The alignment of external and internal pressure postively affect the CE principles (question3).
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python