Basically, what I want to do is, group correlated variables - for e.g. several questions give the end result = internal sources of financing. I would like to have 1 variable made out of these questions as my dependent variable.
Does anyone know how I can do this?
Dear Ivana Jovanovska
Can you explain more about your research question?
Hello Zahra Lashgari
I have a survey data and what I want is to basically group variables that are describing the same effect. For e.g. I have 4 questions that have an end effect = General knowledge of crowdfudning. So, instead of taking all 4, I would like to have them kind of "grouped" (and I was told just a mean value is not a good idea). And I also have more variables that I want to group, in total being 7 groups consisted of several questions from the survey, 5 being independent variables and 2 groups of dependent variables. At the end, I want to do a regression, with my research question being are women choosing internal or external sources of financing before they would choose crowdfunding as a method of financing a business.
I was looking into PCA and factor analysis, but somehow I think this would decrease the importance of the results and it is quite complicated to do with so much groups of variables.
I think you should go through factor analysis. That will help you to define those variables that can be grouped together to be taken further as your dependent variable. In our paper we used oblimin rotation with Kaiser normalisation and dropped items that loaded less than 0.4 or showed less than a 0.10 difference in loadings on more than 1 factor. It's not difficult to do in any of the statistical software (SPSS or Stata).
Dear Ivana Jovanovska
Hello
In my opinion, you may use the Structural Equation Modeling(SEM) and relate your questions to each variable. Using LISREL software, which you can use very easily, if you are not familiar with it, you can see how to use it on YouTube. And with the right input, you will achieve good results.
Good Luck
Yeah, you may use the Structural Equation Modeling(SEM) and relate your questions to each variable. Using LISREL software, which you can use very easily, if you are not familiar with it, you can see how to use it on YouTube. And with the right input, you will achieve good results
Compute the survey questions through Spss by transformations then run normality tests to see if nothing changes.
Rename your variable then you get 1 variable for all your questions
LISREL software can help you in this matter. or you can use SPSS instead by gathering the two measurements questions of the survey in one column by taking an average of those two columns in one column .then run your test.
Hi everyone, thank you for your recommendations.
Factor analysis was mentioned to me already by my mentor. As well as PCA. Since I am doing my analysis in R (I don’t have SPSS on my computer) I had difficulties when trying to do PCA. On top of it, I did a correlation and my main independent variable is not (very weakly) correlated to my dependent variable.
To be honest, my main concerns are the following: I have 26 variables in total. 4 out of those need to be grouped as one - meaning 2 dependent variables; regardining the independent variables, I have 4 that need to be grouped in one variable; and other 5 groups with 2 questions each, with total of 16 variables at the end.
Concern 1: limited time - need to hand in my thesis;
Concern 2: a lot of grouping variables, would the infleunce of the individual variables be significantly reduced …
Concern 3: does it really make sense doing a regression analysis, when there‘a a clear weak relationship between dependent and independent variables …
Any insight would be deeply appreciated.
Use a simple regression analysis to drop some of your independent variables that have low explanatory power. This is the fastest way to drop many of your independent variables and limit the number to be used to two or three maxium in each regression.
Hi Ako Doffou ,
Thanks for your comment. The problem is, I don't want to exclude any variables at this point. I want to see the result with all the variables I already have - this is basically the point of my research. And also, I read that I cannot do a linear regression with Likert scale questions. This is the reason why I am looking for an alternative approach.
Let's take it step by step:
(1) Do not try to compare more than two elements at a time. Your initial analysis should only be on two elements, A and B.
(2) So your initial comparison should only be aimed at establishing the relationship between the answers given to one questiopn and the answers given to another question.
(3) Your next step is to ensure that all the answers receivded to Question A, and all the answers given to question B, are given a quantitaive, numerical, value.
(4) At this point you can either average out the answers to Question A, and the answers given to Question B, and simply limit yourself to a comparative exercise between the two resulting averages.
(5) An alternative could be to give each answer received to Question A, and each answer given to Question B, a range value, e.g. from i to 10.
(6) For either of steps 4 and 5 above you can construct a graph, with the vertical axis representing the range of averages or values resulting applicable to A, and the horizontal axis representing the range of answers applicable to B.
(7) Plot a curve across such axes. The outcome will give you a fair idea of the relationships between the two variables.
As said this is an exercise limited to two questions. Youi can follow the same exercise for three or more questions or groups of questions.
Hope to have been of some help.
Dr. John Consiglio
Faculty of Economics, Management and Accountancy,
The University of Malta
(
Just take the weighted average of the two survey variables.
Problem solved!
- Badges
- Science topic
- Similar topics
- Statistics
- Survey