I would like to measure the causal effects among variables in a path model, but the variables (time series) may be serially correlated. I'm looking for a method to remove or otherwise manage the serial correlation in the data.
Could you provide us with more information?
Sometimes the correlations can be 'removed' by taking the differences between moments instead of analyzing the moments themselves.
Another way could be to model the correlations as well. This could be done using SEM and estimated the correlations. You have to be very careful, because you will need restrictions to have a identified model. You might have more parameters to estimate then your data allows.
You could also look at miultilevel modelling, but if this is an option depends on the number of measurement moments.
Dear Fabio Lepori ,
You can perform generalized differencing, Cochrane-Orcutt procedure, or Hildreth-Lu procedure.
You can remove serial correlation from a regression model by using first differencing method without including the intercept (c) in Eviews software and this is called AUTOCORRELATION. After this, you will recheck through Correlogram-Q-Statistics and Serial Correlation LM Test., and all the probability values after removing serial correlation must greater than 5% level of significance
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series