- This topic has 8 replies, 2 voices, and was last updated 9 years ago by
.
-
Posts
-
Hi Dear,
Actually, i am doing quantitative research (secondary data). Base of some tests i came to know that my data (variables) is not normal. I referred to the book which is mentioned to transformation data by (Log) for making normality of data. I want to ask, is it correct to logarithms of data before analyzing of data? I will be appreciated for your kind guidance.
depends on the kind of analysis you’re trying to do it may (or may not) make sense to transform your data… if you could tell us a little bit more about your data, what you’re looking for (i.e. research hypothesis) and the specific statistical technique you’re intersted in using i think we can all give you a better answer…
Hi Oscar,
Thank you so much for your respond. My area research is accounting and resource of data is financial reports of companies. My data is panel data (time series cross-section) and i am applying multiple regression and moderated multiple regression as statistic methods. Thanks in advance.
My area research is accounting and resource of data is financial reports of companies. My data is panel data (time series cross-section) and i am applying multiple regression and moderated multiple regression as statistic methods. Thanks in advance.
uhm… well, there are a couple of issues here. financial data rarely follows the normal distribution. although i know there are a standard set of tricks to deal with it (such as taking the logarithm of income) please keep in mind that not all data is amenable to such transformations (counts, for instance) and the interpretation of the regression coefficients changes.
what concerns me somewhat is that you’re dealing with data that sounds may (or may not) be serially dependent with some sort of autocorrelation structure… have you done the appropriate tests to make sure that you can ensure the independence of your observations? or how many waves of data do you have on the time series part?
Well, you are completely right. But the issue is I can say that all of my variables is not normal even most of their Mean are less than their standard deviation. I just want to make them nearly normal that I can run statistic test. And other issue is when I transform them by Log, the negative data is ignored (some of my variables include negetive and positive data. I was discussing with a person who is working on time series data and He was saying that this is because of fluctuation of data which is coming from economic crisis. By the way the time series of my data is 10 years (2000-2010 ) from 136 companies.
Thanks again
Well, you are completely right. But the issue is I can say that all of my variables is not normal even most of their Mean are less than their standard deviation. I just want to make them nearly normal that I can run statistic test. And other issue is when I transform them by Log, the negative data is ignored (some of my variables include negetive and positive data. I was discussing with a person who is working on time series data and He was saying that this is because of fluctuation of data which is coming from economic crisis. By the way the time series of my data is 10 years (2000-2010 ) from 136 companies.
Thanks again
well…. if you must then the standard practice would be to add a constant to all your variables to make them positive (so that’s going to affect your interpretation of the intercept) and then do the logarithmic transformation (which will also change your interpretation of the regression weights)..
… betwee you and me, i think you should try to look for some expert help with this. your data is nested at multilple levels (across companies and across time) and i know for a very well fact that skewed financial data is modeled considerably better through something like gamma regression but, alas, i only took a couple of courses in econometrics and stochastic finance in college so i tend not to deal with the peculiarities of your kind of data very often. nevertheless, my intuition tells me that a simple multiple regression analysis is going to have problems… i mean, you’re already starting to get into problems with all the transformations you need to do. you should really consider just ditching regression altogether and favour a more sophisticated technique that can handle the data that you have….
- You must be logged in to reply to this topic.
