I examine the performance of foreign-owned firms in comparison with domestic-owned firms in several profitability indicators (e.g. gross profit/sales and EBIT/sales) for the last 10 years. I plan to run a paired-t test, is it applicable for panel data? Thank you!!
I will have to know a bit more about the nature of data. For example, do you have time-series data for domestic and foreign firms, i.e., panel data? In any case, since you have other factors and covariates, I don't think you would be using the paired-t-test for your problem. I can help you once I know a bit more about your data.
No. Paired t-test would not be appropriate because you don't have anything paired, if I understand your comment correctly. A t-test with measure of success as dependent and domestic/foreign as your independent would be better.
A different approach may be to use mean measure of success as dependent and regress on year using a regression model for both domestic and foreign. That way if the intercept is different you know that one is generally more successful, if the slope is different you know that growth of success is different, and if the lines cross you know that one type starts slow but is ultimately more successful. Purists may disapprove, but this will visually convey your message and that is more important to tell your story. There are more advanced ways of approaching this using more complicated regression models.
No, t-test won't be applicable if there are covariates present in the data.
Thank you so much for all of your comments..
A bit more about my data: I have 1000 Indonesian companies in my samples, consist of approximately 500 foreign-owned companies and 500 domestic-owned companies. Using period of study 2004-2013, I want to test whether there are significant differences between the foreign-owned companies and the domestic-owned companies in several profitability indicators such as: gross profit/sales and earning before interest and taxes/sales.
The independent sample t-test. Specifically the two sample t-test. You should carry out the two independent sample t-test separately for all variables in interest .
I agree with both Chase and Juehui, if you have monthly data for a large number of foreign firms and domestic funds then use regression with a dummy variable for OWNERSHIP (foreign or domestic) as one predictor, time as a continuous predictor (you may need to use a polynomial trend if time series plot suggests that), see if residuals are auto-correlated, and if so, use an ARIMA model with all the predictors. You can use R to do all of this. The site www.stats24x7.com has examples of Time Series Regression.
Thank you very much, everyone! Using STATA, I run regression for each indicator with a dummy variable for ownership as the explanatory variable. I also plan to use propensity score matching to further test the differences. Using natural logarithms of total assets (logAssets), as the independent variables, I match each observation from foreign-owned to an observation of domestic-owned to find to the nearest neighbour propensity score. I have yearly data. Do you have any suggestions for this?
For accounting data such as profitability, it is quite common to use median rather mean e.g. Wilcoxon signed rank test. If you just want to compare/contrast profitability over 10 years, the simple approach may be to run the rank test for each year if you interested with the annual difference or run the test with average data if you are only interested with 10-year average data.
Is the data set an annual series of length 10, for several firms, some of which are domestic and some are foreign? If so, you should not use the paired t-test or Wilcoxon rand test both of which which assume that the n ;pairs' of data are independent, but the data most probably is auto-correlated. I would use panel data analysis with Domestic-Foreign information as a dummy variable.
Hey, Singh and Shi, I have a similar stituation with Purta. I am testing the differences of the board characteristics (such as board size and board independence) between state-owned firms and non-state-owned firms. I use a panel data from 2008 to 2013. So I think ur ideas are great. I would also like to do the panel analysis. But I also have a question: should I control the time variable "year"? It seems necessary, but i am also not sure...
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size