How to compare two different surveys in order to look for a correlation?

[Kiara]

Hi,

I am very confused and I hope there are some smart brains out there to help me understand a bit of statistics

I am writing my dissertation atm and I have collected all my data. I have one survey about emotions and another on organizational commitment. My aim is to compare these two and see if there is a relationship between them. Are emotions increasing the organizational commitment?

However, the survey on emotions have 150 questions, and the survey on organizational commitment have only 21.. I have not made them myself, they are taken from the literature.. Both of the surveys are measured in a 7-points Likert-scale.

As far as I know correlation and regression analysis are the ones to use, but how? Do I need to have independent and dependent variables? Do I take out certain questions and compare them to each other or?

My problem is that I don't know which variables to look at.. Can anyone help?

[victorxstc]

Hi dear Kiara, and welcome

As a suggestion, I think you can simply check the correlation between the total scores of the two tests. This can show you whether by increasing the total score of emotion, the score of "organizational commitment" increases? It needs a simple correlation test.

Also it is possible to check the correlation between each question of each group with each question of the other group. Here you can have 21 x 150 tests which can be simple correlation, but I suggest that you can set each score of the 21 "organizational commitment" questions as a dependent variable in an ordinal regression model, and put the scores of the 150 "emotional" questions simultaneously as its independent variable.

But I would personally run the first test (a simple Spearman coefficient) because the latter will become too cluttering IMHO.

[GretaGarbo]

How many organisations did you investigate and how many interviews did you do on average within each organisation?

What level is this for? Is it for a PhD-dissertation?

[Kiara]

Thank you for tips, victorxstc. I will try to do what you suggested (Do I import both questionnaire in SPSS or do I generate the total scores separately and then work out the rest?) I am not that well known with SPSS, but thank God for Google

PMGretaGarbo: I have only investigated one organization, but across the world. What I wanted to do was to compare all the different teams(countries/regions) with each other. So all the managers in one team has taken the emotional questionnaire, while all the subordinates has taken the organizational commitment questionnaire. However, i am afraid it is too little data on each team? For instance, one of the teams has one manger with 3 subordinates.. In total I have 12 managers and 38 subordinates..

[victorxstc]

You are so welcome

(Do I import both questionnaire in SPSS or do I generate the total scores separately and then work out the rest?) I am not that well known with SPSS, but thank God for Google

Well both can be fed into SPSS together. For example a SPSS file with 150 columns for scores of "emotional" questionnaire, + 21 columns for "organizational commitment (OC)" questions, + 1 column for total emotional scores, and + 1 another column for total "OC" total scores. I think you well know that each row is for each subject surveyed.

However that is not a big deal. You can create two separate files (one dealing with the total scores and another dealing with all scores) as well.

[GretaGarbo]

Is it the same people that answered both questionnaires?

Can you match together these to files so that you one individuals answer for both questionnaires “connected” in the same file?

Is this for a PhD dissertation?

[Kiara]

PMGretaGarbo: No, it is not the same people who have taken the separate questionnaires. Managers has taken the emotions questionnaires, while their subordinates has taken the organizational commitment questionnaire.

No, this is a master dissertation.

[GretaGarbo]

@victorxstc and @kiara,

For each of the 12 mangers (answered 150 emotion questions) there are a few “subordinates”, like 2, 3 or 4 or something. In total 38 “subordinates” (answer 21 organisational commitment questions).

But how are you going to put that together in one file?

There isn’t any direct match between mangers and “subordinates”. How are you going to calculate the correlations?

It would have been nice to hear how kiaras supervisor and teacher had planned to do or evaluated this? I just really would like to know. I am just curious how science works in practice, what was thought in advance and what was not? There are always limitations you know kiara. That is learning to.

One crude way could be to create a table of 12 rows by 150 columns/variables and then for each question (out of the 21) create one column per “subordinate”. That would give 12 rows and 38 times 21 columns that can be matched with the manager data.

The problem is that there will be different number of columns since the manager will have different number of “subordinates”. Say manger 1 have 2 “subs” that gives 2 columns. But if manager 2 have 4 “subs”, that will give 4 columns.




(It feels like I am writing to many posts now. Maybe I should cut down on this.)

[victorxstc]

But how are you going to put that together in one file?

Thank you dear GretaGarbo

Actually I think it is quite simple to put all those data into one single SPSS file. Actually I have made files nearly as this big, and the resulted large file was convenient. Si I personally think it might be possible.

Edited:

(It feels like I am writing to many posts now. Maybe I should cut down on this.)

I, for one, enjoy (and learn from) your valuable hints.

[Kiara]

Actually my supervisor never said anything about it, she said it sounds good (?).. however, now I am scratching myself in the head bc I have no idea how to go through with it and she is on holiday.. yey!

But, to explain a bit more.. I am "measuring" the level of the manager`s emotional intelligence and I want to see if the level of intelligence(or competency) has an effect on the subordinates org.commitment.. so, I was hoping that I somehow can (with limitations) state the level of emotional intelligence/compentency, and compare it to the subordinates? and then see if there is a pattern or similarities among the other teams..

Maybe I should categorize the emotions questions and compare each category to org.com?

I really appreciate your response, @GretaGarco and @Victorxstc !

[noetsi]

You might want to look at the structual equation literature (or scale development) which deals with correlations between variables

[victorxstc]

You are so welcome dear kiara

Yes categorizing the questions and summing up the total scores for each category is another good idea. Besides, since you are working with EQ which is something definable as a single score, perhaps assessing the correlations between the OC scores (total, or each question, or some subgroups) with total emotional scores (the EQ) can be fine.

[Kiara]

When I try to sum up the total score I don't get any scores up.. Can I not sum the total score as long as my value of questions are in the nominal measure?

[victorxstc]

I think you can sum up them in Excel or OpenOffic Calc, by simply copying your data from the SPSS file into Excel and then running a small Excel function to sum them up.

We thought that there might be at least two possible solutions:

1. Since each director influences the subordinates in his/her department, therefore a correct way is to aggregate the results of all the employees in each department and calculate their average score in each category of commitment sub-scores (and also in its total score) and put the subordinates' average value in the same row where we have put the EQ of the manager in that specific department. This methods seems correct but it will cut down your sample size to 12 rows (as for 12 departments).

2. The second suggested solution was that the EQ of a manager affect the behavior (commitment level) of all his/her subordinates in his/her department. So we can assume that each subordinate has actually one manager. Now in this model, each subordinate has a manager who happens to be the manager of all the other subordinates in that department. So if we agree that this model is correct, we can put all the subordinates (without calculating their average values) in rows of our file. OK, in each department there are for example 4 employees and 1 manager. Now we have 4 rows in our SPSS file for the employees of that specific department, but only one row has now a manager EQ score. The point is, since the manager actually manages every single subordinate, his/her EQ score can be put in the row pertaining to the commitment score of each subordinate. So this way we can copy the EQ score of the manager of each department into the empty rows in the EQ column in the rows reserved for that department. This way the sample size will not be reduced.

But we should first make sure that this second method is not considered data fabrication. maybe we can say that our unit of measuring sample size in this particular study is not the number of individuals present in the departments. Instead, perhaps our unit of counting the sample size is the emotional link between the managers and the subordinates. So if we have a manager and four subordinates in one department, we can say the manager is actually influencing all those four subordinates. So in this department, our sample size is not 1 + 4 = 5 individuals, but instead, it is 1 x 4 = 4 manager-subordinate emotional links.

(And if we had 2 manager for 4 employees I think we had to say we have 8 manager-subordinate links in that departments (2 x 4) and in that case we had to not only copy the EQ of the managers, but also copy the commitment scores of the employees to put them once besides one of the managers, and another time besides the second manager at that department.) But that was an example only.

[GretaGarbo]

Can I not sum the total score as long as my value of questions are in the nominal measure?

Yes, I believe that it is so in spss. Just change the variables “classification” from nominal to scale. You can do that in the data editor/variable view/Measure column.

An other variable can also be created with a syntax like (if you have old variables like a, b and c):

COMPUTE new_var=a +b+c .
EXECUTE.