In having categorical dependent variable irrespective of the explanatory variables people say linear regression is not appropriate. But some people say using the scores makes it possible to use linear regression. I am a little bit confused I need clarification.
You need to use some form of logistic regression. For details see the attached reference. Best wishes, David Booth
When your dependent variable is coded 0 and 1 then it is fit for logistic regression analysis NOT linear regression analysis.
Generally, the categorical dependent variable is modeled through some form of generalized linear models.
Yes, it is possible to use linear regression on scores. But the score here is in a statistical sense. I think it is probably not the score you had in mind. In a nutshell, the score here is the gradient of the log-likelihood. Instead of using various link functions in generalized linear models, you can do a regression using the gradients of the individual log-likelihood. It might have some computational advantages in old days. But I don't think many people still do this.
@Xiang, yes you are right. In my sense score is meant to be calculated from the categories in a likert scale. But now I came to understand that logistics regression is better.
Thanx
After coding the dependent variable properly logistic regression can be used.
If you have several items (questions) on a Likert scale, you may calculate a score based on the items (first do a Cronbach Alpha on the items to see whether some of the items may have to be recoded). Usually the score is calculated as the average of all the items. The score may then be treated as approximately continuous and can be used as a dependent variable in linear regression.
The short answer is yes. A common example is probit analysis in insecticide toxicity tests where the response is alive or dead. Another case would be in predicting behavior when people choose to buy or not buy a particular stock in the stock market.
You might consider recoding the data. Are the categorical variables actually categorical or are they simplifications of continuous data? I code income as 60k. However, actual income is a continuous variable and there is a logical arrangement of these categories. In contrast color of cars might be categories like red, blue, white, grey, green, yellow, other. Here there is no logical arrangement and a linear regression type analysis would seem pointless.
@Ebert, in my case I have used likert scale to collect response about their attitude which makes the usage of linear regression pointless. Thanks any way.
is it a single statement? or multiple statements? if it is a single statement variable u have to use ordered logistic regression. If you have multiple items under a single variable u can sum up the scores and use linear regression. If the variable is a single statement Likert Scale item it is not possible to use logistic regression, unless u convert it to binary response.
A single Likert-type item may be used as a response variable in ordinal logistic regression.
If your dependent variable is nominal with two categories use logistic regression, if the dependent variable is nominal with three or more categories use multinomial regression, or if your dependent variable is ordinal use ordinal regression. Of course the respective assupmtions of each test should be followed.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts