Stata/SPSS

Ostrom

Please bear with me, I am a lowly sociologist and figured this would be the best place to go

For those of you who use stata - I am just beginning to use it, and having gotten over the syntax hurdle, am finding it much nicer to use than SPSS. Our department doesn't have stata, and I dont have much chance of a successful request. I understand there is a way to navigate SPSS using syntax, although I haven't tried to (and wont be able to as I am working from another college for the next few weeks - the computer I am using just has stata).

For those of you that might have some familiarity with both, do you know if the syntax tends to be consistent across the two? I'm working with patchy historical datasets and the ease of subtle adjustment with the syntax is particularly useful.

Depending on what you are doing, R might help. Its free and there are loads of guides on the internet, as well as loads of packages to add on.

For time series work, R > Stata although there is a steep initial learning curve.

www.r-project.org

Time Magazine

efla wrote: »

For those of you that might have some familiarity with both, do you know if the syntax tends to be consistent across the two?

Not really. They both have the feel of an old-school computer language, but it's definitely not a copy and paste job. Writing a set of commands in one language will provide a general framework for a copy and paste job, but you'll have to adapt each line to fit.

As far as I know, you're a sociology student so I'm going to be appropriately condescending and presume you're dealing with a relatively small dataset :pac:. By that I mean you don't need a super powerful version of Stata. In that case, you might be interested in purchasing Stata yourself on a student package, for the very reasonable price of about €60 for a one-year licence of €120 for a perpetual licence. You can even buy "Small Stata" (but that limits your n to 1200) for €35.

Zaraba is right to suggest R. It's free, widely used among statisticians and a bit more powerful for number crunching than either of Stata or SPSS.

Flamed Diving

Deleted User wrote: »

Depending on what you are doing, R might help. Its free and there are loads of guides on the internet, as well as loads of packages to add on.

For time series work, R > Stata although there is a steep initial learning curve.

www.r-project.org

Could you point me to a decent guide?

Google for 'Farnsworths guide to econometrics in r'

its here:
http://cran.r-project.org/doc/contrib/Farnsworth-EconometricsInR.pdf

good intro which gets you used to the specifics.

There are some amazing financial packages available on the net which are really cool if you have the data to use them on.

Flamed Diving

Deleted User wrote: »

its here:
http://cran.r-project.org/doc/contrib/Farnsworth-EconometricsInR.pdf

good intro which gets you used to the specifics.

There are some amazing financial packages available on the net which are really cool if you have the data to use them on.

Farnsworth...



I love this package already!

Yeah, that is the only reason I can ever remember who wrote it!

Ostrom

Time Magazine wrote:

As far as I know, you're a sociology student so I'm going to be appropriately condescending and presume you're dealing with a relatively small dataset :pac:. By that I mean you don't need a super powerful version of Stata. In that case, you might be interested in purchasing Stata yourself on a student package, for the very reasonable price of about €60 for a one-year licence of €120 for a perpetual licence. You can even buy "Small Stata" (but that limits your n to 1200) for €35.

Zaraba is right to suggest R. It's free, widely used among statisticians and a bit more powerful for number crunching than either of Stata or SPSS.

The datasets are countrywide - demography and agriculture. Quite large with a small few categorical variables. I'm planning to spend a few months on them (its for my phd) so am willing to invest the time in the learning curve. Categoricals are important so mostly logistic regression. I have a few small sample case studies alright but not sure what to do with them yet. Am expectings around 700 cases overall. Fuppin sociologists etc Thanks for the advice

Stata and R will do logit models quite easily. I've done a lot of logit modelling in stata, its easy enough to run. I wouldn't shell out for Stata-SE just for logits before trying R though.

If you want to really get to grips with the maths (and issues) with them, then Greene's Econometric Analysis and Cameron and Trivedi's Microeconometrics give quite nice and not that technical (for this type of work) overviews and discussion about logit modelling.

Ostrom

Deleted User wrote: »

Stata and R will do logit models quite easily. I've done a lot of logit modelling in stata, its easy enough to run. I wouldn't shell out for Stata-SE just for logits before trying R though.

If you want to really get to grips with the maths (and issues) with them, then Greene's Econometric Analysis and Cameron and Trivedi's Microeconometrics give quite nice and not that technical (for this type of work) overviews and discussion about logit modelling.

Thanks - if small stata is that price though, i'll probably go with it. I'll check out the books too, could do with going over the maths again.

I don't know if you are going to venture to the world of multinomial logits, but really, really study the problem of Independence of Irrelevant Alternatives.

And if you want any help just post here, there are lots of people with good knowledge of econometrics who are normally very willing to help

Ostrom

Deleted User wrote: »

I don't know if you are going to venture to the world of multinomial logits, but really, really study the problem of Independence of Irrelevant Alternatives.

And if you want any help just post here, there are lots of people with good knowledge of econometrics who are normally very willing to help

Thanks, i'm sure i'll be back soon! I may end up there yet when i have a look at whats available in the other datasets.

Flamed Diving

This book is great:

Regression models for categorical and limited dependent variables, J. Scott Long (1997)

Ostrom

Rather than starting a new thread -

I'm still flitting between SPSS and stata - anyone know offhand if there is an SPSS syntax equivalent of stata's */ for inserting notes into your SPSS do file?

Time Magazine

Click on File > I am stupid so I use SPSS > New > Totally talk it over section

Just start with an asterix and finish with a full stop

Ostrom

Time Magazine wrote: »

Click on File > I am stupid so I use SPSS > New > Totally talk it over section

Just start with an asterix and finish with a full stop

Unfortunately I dont have a choice in the matter. For now....

And I am feeling a little more stupid than usual today.

Thanks!

BESman

I like the pretty pictures SPSS produces.

chave

hi,

I would like to test my panel data for autocorrelation and heteroskedasticty?

I know the DW and Breusch Pagan are the tests to do this but I cant find out
how to get stata to do it for panel data.

i used xtreg . . . , fe to do my regressions.

thanks

Time Magazine

chave wrote: »

I would like to test my panel data for autocorrelation and heteroskedasticty?

http://www.stata.com/support/faqs/stat/panel.html

Flamed Diving

What frequency are your panels?

Have you tried "hettest"?

and "estat dwatson"?

Try xtgls, which is Generalised Least Squares for panel data models:

xtgls fits cross-sectional time-series linear models using feasible generalized least squares. This command allows estimation in the presence of AR(1) autocorrelation within panels and cross-sectional correlation and heteroskedasticity across panels.

chave

Flamed Diving wrote: »

What frequency are your panels?

Have you tried "hettest"?

and "estat dwatson"?

Try xtgls, which is Generalised Least Squares for panel data models:

xtgls fits cross-sectional time-series linear models using feasible generalized least squares. This command allows estimation in the presence of AR(1) autocorrelation within panels and cross-sectional correlation and heteroskedasticity across panels.

Im dont really know what im doing, basically did a hausman and it said use random effects , I just need the commands to do the tests for Heteroskedasticity and Autocorrelation within a RE model???? Ive read in a book that they differ from the FE tests.

thanks

Flamed Diving

chave wrote: »

Im dont really know what im doing, basically did a hausman and it said use random effects , I just need the commands to do the tests for Heteroskedasticity and Autocorrelation within a RE model???? Ive read in a book that they differ from the FE tests.

thanks

Ok, I know little of FE and RE, but...

A recommendation in Econometric Analysis of Cross-Section & Panel Data (Woolridge, 2002), suggests that, under certain assumptions, the Langrange multiplier statistic in a likeihood setting, developed by Breusch and Pagan (1980) will detect serial correlation. The book was more or less silent on heteroscedasticity tests or dealing with it.

Ok, so then I referenced Econometric Analysis (Greene, 2002). On page 316, section 13.7.2, he discusses possible ways to deal with it, but actually arrives at a indeterminate conclusion. He does mention that using GLS will correct for this (as it always does) and doing a little jig with the regression will get you FGLS, if that's what you want. There is a very recent journal paper in Econometrica which is still aiming to solve this problem, so my guess is that it hasn't been resolved yet. Greene's book is the bible in econometrics, so if he doesn't have the answer, no one does.

I also found this suggestion:

http://www.stata.com/statalist/archive/2005-01/msg00330.html

Which isn't a terrible idea, and bootstrapping your SE's is just a case of typing 'bootstrap'. But no test.

chave

Flamed Diving wrote: »

Ok, I know little of FE and RE, but...

A recommendation in Econometric Analysis of Cross-Section & Panel Data (Woolridge, 2002), suggests that, under certain assumptions, the Langrange multiplier statistic in a likeihood setting, developed by Breusch and Pagan (1980) will detect serial correlation. The book was more or less silent on heteroscedasticity tests or dealing with it.

Ok, so then I referenced Econometric Analysis (Greene, 2002). On page 316, section 13.7.2, he discusses possible ways to deal with it, but actually arrives at a indeterminate conclusion. He does mention that using GLS will correct for this (as it always does) and doing a little jig with the regression will get you FGLS, if that's what you want. There is a very recent journal paper in Econometrica which is still aiming to solve this problem, so my guess is that it hasn't been resolved yet. Greene's book is the bible in econometrics, so if he doesn't have the answer, no one does.

I also found this suggestion:

http://www.stata.com/statalist/archive/2005-01/msg00330.html

Which isn't a terrible idea, and bootstrapping your SE's is just a case of typing 'bootstrap'. But no test.

thanks man

Ostrom

Anyone know if stata press books are sold in any Dublin bookshops? Or Uni bookshops - I'm within range of trinity and UCD

Time Magazine

efla wrote: »

Anyone know if stata press books are sold in any Dublin bookshops? Or Uni bookshops - I'm within range of trinity and UCD

Most decent Stata press books (as in the ones by Cameron and Trivedi, Baum, etc.) are available at reasonable prices online. The manuals are crap imho, just use the website.

You might also grow a bit of a brass neck and email some economists. Most people have books just sitting on their shelves and they wouldn't mind lending them to a respectable PhD student who would benefit from borrowing it for a month.

You can PM me the book(s) you need and I'll see what I can source for you, if you want.

Snackie

SPSS is good for data management, but you should run regressions on Stata