Dear Mahmoud:
First, you ask for a section on guidelines/rules about using Stata -- I
think you mean Statalist. This is precisely one of the aims of the FAQ,
which in a previous email you say you have read.
You, and indeed anybody else who wants some other document, should
please explain what they want, because it is not clear to me. As FAQ
maintainer, I have a strong bias against providing _two_ documents and
against replicating suggestions already given. It will be not obvious to
new members, but we have had a FAQ for most of Statalist's 14-year-old
history and have tried various ways of distributing it. I may seem
conservative on this point but I am clear that the present method of a
FAQ on a website is the best way to make information on Statalist
publicly and permanently accessible.
Of course, that in no way inhibits or prohibits anybody else from
writing what they want, from posting it to the list or putting it
anywhere on the web. That would be their personal view, but perhaps
illuminating or informative to others.
Second, the t-test you use is, like the bootstrap, predicated on
independence of observations. Thus any P-values and c.i.s you get will
be way off if your data show serial correlation. As you have time
series, that would very likely be the case. There is scope for
approximate correction, but the -ttest- command you use does not in any
way take account of any time series structure in the data.
This is well explained in (e.g.) Box, G.E.P., Hunter, J.S., Hunter, W.G.
2005. Statistics for experimenters. Hoboken, NJ: John Wiley.
The first edition would also be suitable.
This book is also a good one for basic statistics in general.
Third, you, or your local systems person, should -update- from Stata 9
to 9.2. It's free.
Nick
[email protected]
Mahmoud Abd-El-Aal
[...]
I do definitely support a section
about guidelines/rules about using stata, then it would be much easier
to
know what is considered appropriate behaviour
[...]
I am using stata version 9
in order to compare two variables means (how many of the means of the
bootstraped samples are larger than the original series sample) i had
the
following commands in my mind:
gen mean diff= var2-var1
ttest diff==0
One-sample t test
Variable Obs Mean Std. Err. Std. Dev. [95% Conf.
Interval]
diff 3516 .0006044 .0000962 .0057026 .0004158
.0007929
mean = mean(diff) t = 6.2842
Ho: mean = 0 degrees of freedom = 3515
Ha: mean < 0 Ha: mean != 0 Ha: mean > 0
Pr(T < t) = 1.0000 Pr(T > t) = 0.0000 Pr(T > t) = 0.0000
as i am new stata just see if this interpretation is correct, so we
reject
the null hypothesis of mean(diff) is 0 so its either bigger or smaller ,
but from what i see the mean fall with the 95% CI right? another thing
looking at the Ha: mean<0, does that mean that var2(mean)<var1(mean),
[...]
*
* 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/
