[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: -qlognorm- package available on SSC
I am guessing that, despite the advent of -ssc whatsnew- on
2 August 2002, announcements of packages new or substantially revised
on SSC are still of some interest.
Thanks to Kit Baum, a package -qlognorm- containing -qlognorm-
and -plognorm- programs has been added to SSC.
-qlognorm- is a package for diagnostic plots for the
lognormal distribution, somewhat in the spirit of official
Stata's -qnorm-, etc.
-qlognorm- plots the quantiles of varname against the quantiles of the
corresponding lognormal distribution.
-plognorm- graphs a standardized lognormal probability plot for
The (two-parameter) lognormal distribution fitted to varname
corresponds to a
normal distribution with the mean and standard deviation of
Stata 7 is required.
Sometimes there is interest in whether the lognormal is appropriate as
a distribution model for a variable. Other times there is interest in
whether the logarithm of a variable is more nearly normal than that
These are two sides of the same question. -qlognorm- and -plognorm-
are commands for investigating it directly.
With official Stata, it is easy to generate a new variable which is
the logarithm of a variable and then to use -qnorm- and -pnorm- to see
whether that new variable is close to normal in distribution.
Using -qlognorm- and
-plognorm- instead has these small but distinct advantages:
1. If you do this frequently, you will need to type less; sometimes,
but not always, you will decide that a log transformation is
2. Fit can be assessed graphically on both raw and transformed scales.
3. If desired, you can use a plotting position other than the i / (N +
1) wired into -qnorm- and -pnorm-. There is a modest literature on
of plotting positions in probability plots, and some grounds for
positions other than what official Stata has chosen (somewhat
so far as I can tell). For a little more discussion, see
4. If desired, you can insist on maximum likelihood estimation. That
the standard deviation used by default is that emitted by -summarize-,
which is the square root of the variance estimated with (N - 1) as
divisor, which is not the maximum likelihood estimator. This option
is for purists: if the difference between N and (N - 1) makes a
to your results, this should be resolved by a bigger sample, not
by standing on principle. Still, some people like being purists.
* For searches and help try: