Though there are a lot of significant discussion about this issue but today I am confused when I am putting my data into data sheet from the questionnaire.
I have variable name social capital and in my research, I have taken six dimensions of social capital (*e.g. trust, social cohesion etc.) and in every dimension, I have several questions. Data were collected by using 5 points scale (strongly agree to strongly disagree) and in the questions, I have both positive and negative questions equally (e.g. I have trust and I do not have trust).
Now, I am confused. In the case of positive statement - should I have to give 5 for strongly agree and for negative statement should I have to give 1 for strongly agree?
Thank you in advance for your anticipated cooperation.
You need to reverse the items if you are going to calculate Cronbach's alphas, which assumes all the correlations are positive. Beyond that, it would be wise to run at least an Exploratory Factor Analysis, and even better, a Confirmatory Factor Analysis.
For negative statements, you should reverse your scores, unless you have specific reasons for not doing so.
In the first place, it has to do a recoding of the negative (or positive, it is indifferent) propositions. Code 5 becomes 1; 4 becomes 2, 3 stays the same; 2 becomes 4 and 1 becomes 5. In the next step, youust realize a factorial analysis to find the dimensions that correlate the respective indicators. I was clear? Good luck in your research
Based on my understanding, it depends on your statement/questions. If your question is positive, like does this intervention/action enhance social cohesion? in that case strongly agree should get highest score (5); In contrast, does this intervention reduce social cohesion? in such case strongly agree will get lowest score (1), I hope it will help.
You tell I have trust, answer 5, means strongly agree. I don't have trust, answer 1, means strongly disagree, that means you have trust.5 for positive question is same 1 for negative question. In other words, you convert negative question to positive question.
You need to reverse the items if you are going to calculate Cronbach's alphas, which assumes all the correlations are positive. Beyond that, it would be wise to run at least an Exploratory Factor Analysis, and even better, a Confirmatory Factor Analysis.
You say that you are using a 5-point Likert scale from "strongly agree to strongly disagree". Is this:
strongly agree
agree
neither agree nor disagree
disagree
strongly disagree?
In which case, your mid-point 'neither agree nor disagree' might be allowing respondents to 'sit on the fence'; therefore, deleting this mid-point might be worth considering.
First, let me state that I agree with reverse approach as suggested by Siddhi, David L. Morgan and others. However, what you have not told us is how who plan to use the data after the appropriate coding. Often, these issues are agreed upon prior to the data collection by the concerned parties. Is it as a regressand or a regressor. This is where Philp Adams comment warrants further discussion.
Dear Arize, I have some other variables measured in different scale. Therefore, I have both continuous and categorical data. I have two group i.e. control group and treatment group. My focus is to identify the change in the capacity level of the treatment group and to identify the important factors that might influence in this changing process. I had a plan to a Multivariate analysis or ANCOVA. But right now, I am in the middle of the decision.
After using this coding, I have a plant to use the total score of individual into my analysis as a continuous variable.
All of your comments and suggestions will help me to take a smooth decision.
Thank you
The assumptions of ANCOVA can lead you that whether you would be better to coding variables into continuous variables or leave these in the same form as they are. You can take help from the following links.
http://www.real-statistics.com/analysis-of-covariance-ancova/assumptions-ancova/
M. Imran
Since you have continuous plus categorical data, it seems that ANCOVA will be and such will allow likert scale score total for ith individual to be treated as interval data.
ANCOVA is identical to a regression with the treatment specified as a dummy variable. The advantage of the regression is that it would be easier to examine the separate effects of each of the control variables.
David's suggestion concerning analysis of covariance in a regression format is what you you need to do. It is very simple and flexible since you can use differential intercept and differential slope or interaction variables. Further, you can handle any nonspherical errors within that framework.
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Applied Statistics
- Survey Design