Re: st: FW: How can I analyse a timeseries dataset appropriately?

[Gordon Hughes <G.A.Hughes@ed.ac.uk>]

Since no-one else has picked this up, let me offer a few brief comments:

1.  You have panel, not time series, data - 20 
observations on 5 newspapers.  Look at -xtreg- or 
-xtregar- with either fixed (FE) or random 
effects (RE).  There are matters of statistical 
philosophy that differentiate these methods, but 
there might not be a statistically observable 
difference in practice when you have only 5 panels.

2.  You have a set of time series variables that 
are identical for all 5 panels, so that any 
differences between newspapers reflect either the 
editorial policies of the newspapers or some 
other panel-specific variables.  On this basis 
you can't estimate separate time dummies.  You 
should consider getting more variables on the 
characteristics of the newspapers - e.g. the 
average income or education of their readers.

3.  You refer to random coefficient 
models.  Given the nature of your data, -xtrc- 
should be equivalent to seemingly unrelated 
regression -sur-, but again the number of panels 
is small.  You could look at -xtmixed- using newspaper as the id variable.

Gordon Hughes
g.a.hughes@ed.ac.uk


> Date: Thu, 17 Feb 2011 19:33:36 +0100
> From: "Claus D. Hansen" <claushansen@pc.dk>
> Subject: st: FW: How can I analyse a timeseries dataset appropriately?
>
> Dear list-members,
>
> I have a dataset I have assembled myself but I am having trouble figuring
> out just how to analyse it properly.
>
> I have a dataset consisting of a variable indicating how big a share of the
> total articles in a newspaper sickness absence has had for that specific
> year (this is a continuous variable).
>
> I have this information for 5 newspapers in the period 1991-2010 which means
> I have 100 observations all in all.
>
> For each year I have some information ­ i.e. what government is ruling in
> that year (dummy variable), how big unemployment was, the average sickness
> absence rate and how many public money was used on sickness absence (last
> three are all continuous variables) ­ these
>
> I want to know which of the independent variables predict the share of
> articles about sickness absence ­ and I want to know if there is a ?time
> effect?, i.e. I want to know if there is an association between year and the
> share of articles on sickness absence when adjusting for the above mentioned
> independent variables ­ and if the ?time effect? disappears I want to know
> which of the independent variables are ?responsible? for explaining how the
> ?time effect? disappears.
>
> I know that it will be wrong to use standard OLS ­ because it?s a time
> series the observations will not be independent - another problem is that my
> independent variables are not specific to each of the 5 observations on the
> share of sickness absence articles.
>
> I suspect I should be using some kind of random effects modeling but I might
> be wrong ­ can anyone point me to the right way of analyzing this?
>
> Thank you in advance,
>
>
> Claus D. Hansen
> Assistant professor
> Aalborg University
> Denmark


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/