For my thesis I want to assess the effect of imports from china on manufacturing unemployment in 12 OECD countries. For this I use data from Input-Output tables (WIOD). This gives me data on intermediate goods trade from 14 manufacturing industries in China to any other industry in the 12 OECD countries, over the time period 1995-2011. I sum the trade from 1 Chinese manufacturing industry to all industries in for instance Germany in a certain year. The dependent variable will be total hours worked on the industry level. This results in a data set with 14 industries per country, 12 countries and 17 years. How should I analyse the data?
I think there are two options:
-Mixed model (multilevel data with repeated observations). And I think has crossed instead of nested factors, as the 14 industries are the same in each country
-Fixed effects model (panel data), with country (and industry?) dummies. (fixed effects for years and countries (and industries?)). But then the problem might be that I cannot use control variables at the country level?
Could anyone tell me which one I should choose and why?
You may want to read this
https://www.researchgate.net/publication/233756428_Explaining_Fixed_Effects_Random_Effects_Modeling_of_Time-Series_Cross-Sectional_and_Panel_Data
and
https://www.researchgate.net/publication/299604336_Fixed_and_Random_effects_making_an_informed_choice
and on cross classifications
http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#nonhier
and
How many countries for multilevel modeling? A comparison of Frequentist and Bayesian approaches.
https://pdfs.semanticscholar.org/5af9/a5697ee71d6f869a5a5d5887f47dc4bd205c.pdf
Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
Article Fixed and Random effects models: making an informed choice
Hello
As you know each approach and model in econometrics should be selected and employed according with theoretical issues and data availability. So with regard ti this and your explanation you must choose the approach that make sense. I think the last one is better because your section and year are enough to regress and estimate panel data regression. But there is question still open to discuss: "Why do you want to use control variables"?
In my opinion Panel Data is more usual for this kind of data.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Energy