I am trying to analyze the effect of ease of use, security and other factors (independent) on e-payment adoption(dependent), all the variables are in likert scale.I have not done much econometrics in postgrad as I am studying strategic management, can you tell me what type of regression to do and how to go about it ?
You can do a simple regression following the form Y = a + bX with both Y and X showing the scores of each respondent. I am of course assuming that you are doing a questionnaire where the responses are scored according to a Likert scale. For two variables regression you can even use an excel spreadsheet to get your a or b values.
If you have Likert scale data --- that is, several Likert items summed or averaged together to make a scale that has many different final values --- then that data is usually treated as continuous.
You will benefit from reading up on creating Likert scales from a set of Likert items, for example calculating "reliability" measures.
Also, be cautious with the assumptions of common parametric tests (t-test, linear regression) with Likert scale data, as the data may not be marginally normal or homoscedastic, etc.
The issue is that some people say that Likert scales are only rank data, not interval data. Nonetheless it is common to use linear regression and related statistics.
BUT and this is important, do check all your descriptives and make sure each scale is
- normally (or at least uniformly) distributed
If the scale is badly skewed, think of rescaling it.
It will be wise to at least check the reliability of each scale using Cronbach alpha.
And even start one step back and do a principal components analysis and drop any items that don't load on the dominant component.
Grad a psychologist to help you. They are usually pretty good on this.
Likert scale is an established way of measuring latent variables specially in Organizational Studies and is considered continuous form of data. Of course Validity Reliability must be established. So I think Simple Linear Regression in SPSS or SmartPLS will serve the purpose. However, sample size and statistical power issues are of significant importance these days.
There are many methods available, and the best one will depend on the nature of your study and the questions you are trying to answer. A popular method is to analyze responses using variance analysis techniques, such as the Mann Whitney test or the Kruskal Wallis test. To compare the answers between several groups of respondents, one can use the Kruskal Wallis test of the variance.
The chi-square test is a possible approach for analyzing data when responses are grouped into two dichotomous categories. Example: Agreement / disagreement.
For regression, I think it's possible to translate both the dependent variable and indepedent variables into dichotomous variables. And then, you can use logistic regression.
If all the variables are ordered categorical, I suppose you could use ordinal logistic regression, see link below. You might consider dummy-coding your independent variables.
https://stats.idre.ucla.edu/stata/dae/ordered-logistic-regression/
Hi Samudra !
What type of regression can be followed ?
The type of regression you can follow is a multiple regression, where you have one dependent variable and two or more independent variable (Bivariate - one dependent and one independent).
What the solution will mean ?
For example, Considering only two independent variables: ease of use (X1) & security (X2) and dependent variable: e-payment adoption (Y), your equation will become
Y = m1*X1 + m2*X2 + c (where c is a constant & m is partial regression coefficient.
from here, you can deduce that for every 1 unit increase in ease of use, e-payment adoptability will increase or decrease by some units depending upon whether both are positively or negatively related respectively.
How to proceed?
You can use a software called SPSS, which is available over net to download trail version valid for 15 days.
Else, you can use excel sheet and download data analysis tool for it, that shall help you do regression both multiple and bivariate.
Dear Samudra Biswas,
your question is not properly clear,
are the data categorical or not? is this situation for all data or not?
your questionary, how many questions belong to one variable? (if you have this situation, you can extract some factors and . . . , you can use structural equation model SEM)
maybe some clarifications could help to solve your problem.
bests...
To add to a previous post , perhaps as they have indicated you could explore the Logic Regression model to do your calculation.
- Badges
- Science topic