Page 1 of 1
Panel data graphs
Posted: Fri May 21, 2010 2:31 pm
by dalleyne
Hello evryone,
I have a regular panel data set with 13 countries with observations per country from (1991-2000).I am looking at the impact of FDI on growth. I am trying to generate a single graph in which the growth rates for each country is shown as a function of time.How do I do this using the graph procedure.Is this very similar to graphing a single series.
Thanks for your help.
DA
Re: Panel data graphs
Posted: Fri May 21, 2010 5:09 pm
by TomDoan
It will probably look a bit busy, but this shows how to split up a panel data series into separate time series and graph them with the time periods coordinated. Note that you have to use MOVE rather than SET because SET won't let you transfer data across individuals (a protection against lagging into another individual's data).
Code: Select all
cal(panel=10,a) 1991:1
all 13//2000:1
set x = %ran(1.0)
dec vect[series] p(13)
do i=1,13
move x i//1991:1 i//2000:1 p(i) 1991:1
end do i
list iser = p
graph 13
cards iser
Re: Panel data graphs
Posted: Sat May 22, 2010 7:52 pm
by dalleyne
Hello Tom,
Thanks for your quick response.The program worked and I now have the graphs I want.I was also able to use the graph label command to rename the new series p1 p2 p3 etc.
Best regards.
Re: Panel data graphs
Posted: Tue Mar 27, 2012 1:01 pm
by tomcy
Hi
I have a panel data set with GDP and other variables for 14 countries from 1999:1 to 2011:4. I would like to pool the GDP data together for the 14 countries and create a single series from 1999 to 2011 which is the sum of GDP across the 22 countries over time. What is the best way to create and graph the pooled GDP series?
Thanks
Re: Panel data graphs
Posted: Tue Mar 27, 2012 3:07 pm
by moderator
You can use a PANEL instruction with the TSUM option to create a series containing the sum across individuals. For example:
panel(tsum=1.0) test / testsum
The output series will contain the same set of values in each individual, so you could just limit your GRAPH instruction to only graph the range for the first individual. For example:
graph
# testsum 1//1999:1 1//2011:4
Alternatively, you could use the COMPRESS option on PANEL as well. This eliminates the duplicated data that appears by default in individuals 2, 3, and so on. If you do that, there's no need to limit the range on the GRAPH instruction, since the series only contains data in the entries for individual one. For example:
panel(tsum=1.0,compress) test / testsumcomp
graph
# testsumcomp
Regards,
Tom Maycock
Estima
Re: Panel data graphs
Posted: Tue Apr 10, 2012 12:09 pm
by bbatson
Hi,
I am working with an unbalanced panel (4 states and ~22 years) and want to make an unstacked graph for some of my series so I can see individual state's certain variables over time in one figure (i.e. 1 figure output with 4 separate graphs). Also, is it possible to put vertical lines in certain years for only some of the states to reflect exogenous individual shocks?
Thank you,
Brian
Re: Panel data graphs
Posted: Tue Apr 10, 2012 12:24 pm
by moderator
bbatson wrote:Hi,
I am working with an unbalanced panel (4 states and ~22 years) and want to make an unstacked graph for some of my series so I can see individual state's certain variables over time in one figure (i.e. 1 figure output with 4 separate graphs).
It sounds like you are looking for something like:
spgraph(vfield=2,hfield=2)
do indiv=1,4
graph
# x indiv//start indiv//end
end do
spgraph(done)
where "start" and "end" are the desired dates or entry numbers. For example:
spgraph(vfield=2,hfield=2)
do indiv=1,4
graph
# x indiv//1990:1 indiv//2012:1
end do
spgraph(done)
bbatson wrote:Also, is it possible to put vertical lines in certain years for only some of the states to reflect exogenous individual shocks?
Sure, you can create series with 1s in the periods where you want vertical bars, and supply these using the GRID option on your graph instructions. If you wanted to flag the same time periods for each individual, you could just use one series. For example:
set gridser = %period(t)==1999:1.or.%period(t)==2005:1
would create a series with 1s for the two time periods in each individual. You could then do:
spgraph(vfield=2,hfield=2)
do indiv=1,4
graph(grid=gridser)
# x indiv//1990:1 indiv//2012:1
end do
spgraph(done)
If you need to highlight different periods in different individuals, you could still use a common grid series, with additional commands to set the desired periods within each individual, or you could create separate grid series. For the latter, you could just use 4 different names and use 4 graph instructions (rather than the DO loop), or you could keep the loop, and use a VECTOR of SERIES for the different grid series, referencing the appropriate vector element in the loop.
Regards,
Tom