I want to study the trend of a variable (say firms' investment in R&D) across four waves of a large-scale cross-country survey. Nevertheless, firms surveyed in one wave may not be the same of the following one, altohough their characteristics (size, nationality, etc...) are similar.
Am I allowed to make t-tests on means between independent samples from one wave to another and discuss a trend accordingly? Do you know of similar approaches in the business or innovation literature?
Thank you very much,
Marco
These data follow within the conceptual framework of cross-sectional time series data. You should first assess whether there is autocorrelation in your data. That is, that the dependent variable in any one wave is correlated with past values of the dependent variable. If so, you should consider a longitudinal regression model that can adjust for autocorrelation. Such models will provide you estimates of the trend as you desire.
Dear collegue
Dear colleague
you can attain transversal and longitudinal (across time) analysis applying the methodology of radar chart. You can obtain several effects of study by this new accounting methodology by published articles on this web or google scholar. The variables obtained by this methodology are always positive, normalized, independent and objective. Later, indicators can be applied according your need of research.
I would caution against using radar charts. The method is not designed for handing the type of data you are dealing with here. The ideal situation for you is to use longitudinal regression models correcting for auto correlation as has been suggested before. Another way would be take samples across all (perhaps bootstrapping may be more applicable) and ensure the presence or absence of autocorrelation before you begin.
I don't know about radar charts. However, many of the tools for panel analysis only work if you have the same firms in each panel. This is a real problem with your data. If the firm appears in multiple panels, then serial correlation is a big issue. However, if you have a firm that only appears in one panel, many panel analysis techniques will ignore the firm. A lot depends on exactly what you want to test.
Dear Colleague
when a transversal study is made, it is need know, what is the behavior of companies on a panel?. On a panel study, economic behaviors of companies may be due to economic sector as well as locations of companies. I am obtaining as results that companies have a economic behaviors according to their location. The measure of this factor does not enter on conflict with other varialbes, and it can be detected by other different tools to the radar charts, as example, the level of credit on a region.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS