{smcl}
{* 19june2002/5july2004}{...}
{hline}
help for {hi:dpplot}
{hline}
{title:Density probability plots}
{p 8 17 2}{cmd:dpplot}
{it:varname}
[{cmd:if} {it:exp}]
[{cmd:in} {it:range}]
[ {cmd:,}
{cmd:a(}{it:#}{cmd:)}
{cmd:dist(}{it:name}{cmd:)}
{cmd:param(}{it:numlist}{cmd:)}
{cmdab:gen:erate(}{it:newvar1 newvar2}{cmd:)}
{cmd:line(}{it:line_options}{cmd:)}
{it:graph_options}
{cmd:plot(}{it:plot}{cmd:)}
]
{title:Description}
{p 4 4 2}{cmd:dpplot} plots density probability plots for {it:varname}
given a reference distribution, by default normal (Gaussian).
{title:Remarks}
{p 4 4 2}
To establish notation, and to fix ideas with a concrete example: consider an
observed variable {it:Y}, whose distribution we wish to compare with a normally
distributed variable {it:X}. That variable has density function {it:f}({it:X}),
distribution function {it:P = F}({it:X}) and quantile function
{it:X = Q}({it:P}). (The distribution function and the quantile function are inverses
of each other.) Clearly, this notation is fairly general and also covers other
distributions, at least for continuous variables.
{p 4 4 2}
The particular density function {it:f}({it:X} | parameters) most pertinent to
comparison with data for {it:Y} can be computed given values for its
parameters, either estimates from data on {it:Y}, or parameter values chosen
for some other good reason. In the case of a normal distribution, these
parameters would usually be the mean and the standard deviation. Such density
functions are often superimposed on histograms or other graphical displays. In
Stata, {help histogram} has a {cmd:normal} option which adds the normal density
curve corresponding to the mean and standard deviation of the data shown.
{p 4 4 2}
The density function can also be computed indirectly via the quantile function
as {it:f}({it:Q}({it:P})). For example, if {it:P} were 0.5, then
{it:f}({it:Q}(0.5)) would be the density at the median. In practice {it:P} is
calculated as so-called plotting positions {it:p_i} attached to values
{it:y_}({it:i}) of a sample of {it:Y} of size {it:n} which have rank {it:i}:
that is, the {it:y_}({it:i}) are the order statistics {it:y_}(1) <= ... <=
{it:y_}({it:n}). One simple rule uses {it:p}_{it:i} = ({it:i} - 0.5) / {it:n}.
Most other rules follow one of a family ({it:i} - {it:a}) / ({it:n} - 2{it:a} +
1) indexed by {it:a}.
{p 4 4 2}
Plotting both {it:f}({it:X} | parameters) and {it:f}({it:Q}({it:P} =
{it:p_i})), calculated using plotting positions, versus observed {it:Y} gives
two curves. In our example, the first is normal by construction and the second
would be a good estimate of a normal density if {it:Y} were truly normal with
the same parameters. In terms of Stata functions, the two curves are based on
{cmd:normden(}({it:X} - mean) / SD){cmd:)} and
{cmd:normden(invnorm(}{it:p_i}{cmd:))}. The match or mismatch between the
curves allows graphical assessment of goodness or badness of fit. What is more,
we can use experience from comparing frequency distributions, as shown on
histograms, dot plots or other similar displays, in comparing or identifying
location and scale differences, skewness, tail weight, tied values, gaps,
outliers and so forth.
{p 4 4 2}
Such {it:density probability plots} were suggested by Jones and Daly (1995).
See also Jones (2004).
They are best seen as special-purpose plots, like normal quantile plots and
their kin, rather than general-purpose plots, like histograms or dot plots.
{p 4 4 2}
Extending the discussion in Jones and Daly (1995), the advantages (+) and
limitations (-) of these plots include
{p 8 8 2}+1. No choices of binning or origin (cf. histograms, dot plots, etc.)
or of kernel or of degree of smoothing (cf. density estimation) are required.
{p 8 8 2}+2. Some people find them easier to interpret than quantile-quantile
plots.
{p 8 8 2}+3. They work well for a wide range of sample sizes. At the same
time, as with any other method, a sample of at least moderate size is
preferable (one rule of thumb is >= 25).
{p 8 8 2}+4. If {it:X} has bounded support in one or both directions, then this
should be clear on the plot.
{p 8 8 2}-1. Results may be difficult to decipher if observed and reference
distributions differ in modality. For example, if the reference distribution is
unimodal but the observed data hint at bimodality, nevertheless
{it:f}({it:Q}({it:P})) must be
unimodal even though {it:f}({it:Y}) may not be. Similarly, when the reference
distribution is exponential, then {it:f}({it:Q}({it:P})) must be monotone
decreasing whatever the shape of {it:f}({it:Y}).
{p 8 8 2}-2. It may be difficult to discern subtle differences in one or both
tails of the observed and reference distributions.
{p 8 8 2}-3. Comparison is of a curve with a curve: some people argue
that graphical references should where possible be linear (and ideally
horizontal). (A linear reference is a clear advantage of quantile plots.)
{p 8 8 2}-4. There is no simple extension to comparison of two samples
with each other.
{p 4 4 2}
Programmers may wish to inspect the code and add code for other distributions.
If parameters are not estimated, then naturally their values must be supplied:
the order of parameters should seem natural or at least conventional.
{title:Options}
{p 4 8 2}{cmd:a()} specifies a family of plotting positions, as explained
above. The default is 0.5. Choice of {cmd:a} is rarely material unless the
sample size is very small, and then the exercise is moot whatever is done.
{p 4 8 2}{cmd:dist()} specifies a distribution to act as a reference.
The distributions implemented include {cmd:beta}, {cmd:exponential},
{cmd:gamma}, {cmd:Gumbel}, {cmd:lognormal}, {cmd:Weibull} and {cmd:normal},
the last being the default. {cmd:Gaussian} is a synonym for {cmd:normal}.
{p 4 8 2}{cmd:param()} specifies parameter values which give a reference
distribution.
{p 8 8 2}With {cmd:dist(normal)} two parameters may be specified. The
first is the mean and the second is the standard deviation.
{p 8 8 2}With {cmd:dist(Weibull)} two parameters may be specified. The
first is a scale parameter {it:b} and the second a shape parameter {it:c}.
(The density function for a variable {it:x} is thus ({it:c}/{it:b})
({it:x}/{it:b})^({it:c} - 1) exp(-({it:x}/{it:b})^{it:c}).)
{p 8 8 2}With {cmd:dist(lognormal)} two parameters may be specified. The
first is the mean of logged values and the second is the standard deviation
of logged values.
{p 8 8 2}With {cmd:dist(gumbel)} two parameters must be specified. The
first is a scale parameter alpha and the second is a location parameter mu.
(The density function for a variable {it:x} is thus
(1 / alpha) * exp[-({it:x} - mu) / alpha] * exp[-exp(-({it:x} - mu) / alpha)].)
{stata ssc desc gumbelfit:gumbelfit} is one program to estimate parameters.
{p 8 8 2}With {cmd:dist(gamma)} two parameters must be specified. The
first is a shape parameter alpha and the second is a scale parameter beta.
(The density function for a variable {it:x} is thus
[1 / (beta^alpha * Gamma(alpha))] {it:x}^(alpha - 1) exp(-{it:x} / beta),
where Gamma() is the gamma function.)
{stata ssc desc gammafit:gammafit} is one program to estimate parameters.
{p 8 8 2}With {cmd:dist(exponential)} one parameter may be specified, namely
the mean.
{p 8 8 2}With {cmd:dist(beta)} two parameters must be specified,
shape parameters alpha and beta.
(The density function for a variable {it:x} is thus
[1 / Beta(alpha, beta)] {it:x}^(alpha - 1) (1 -{it:x})^(beta - 1),
where Beta() is the beta function.)
{stata ssc desc betafit:betafit} is one program to estimate parameters.
{p 4 8 2}
{cmd:generate()} specifies two new variable names to hold
the results of densities estimated from the data directly (as {it:f}() given
parameters) and indirectly (as {it:f}({it:Q}({it:P})) given parameters).
{p 4 8 2}{cmd:line(}{it:line_options}{cmd:)} are options of
{help twoway_mspline:twoway mspline} and {help twoway_line:twoway line},
which may be used to control the rendition of the density function curve.
{p 4 8 2}{it:graph_options} are options of {help twoway}.
{p 4 8 2}{cmd:plot(}{it:plot}{cmd:)} provides a way to add other plots to the
generated graph; see help {help plot_option}.
{title:Examples}
{p 4 8 2}{inp:. dpplot mpg}
{p 4 8 2}{inp:. set obs 1000}{p_end}
{p 4 8 2}{inp:. gen rnd = invnorm(uniform())}{p_end}
{p 4 8 2}{inp:. dpplot rnd, param(0 1)}{p_end}
{p 4 8 2}{inp:. dpplot rnd, param(0 1) plot(histogram rnd, bcolor(none) width(0.2))}
{p 4 8 2}{inp:. dpplot length, dist(lognormal) gen(density1 density2)}
{p 4 8 2}{inp:. gammafit length}{p_end}
{p 4 8 2}{inp:. dpplot length, dist(gamma) param(`e(alpha)' `e(beta)')}
{title:Author}
{p 4 4 2}Nicholas J. Cox, University of Durham, U.K.{break}
n.j.cox@durham.ac.uk
{title:Acknowledgements}
{p 4 4 2}Tim Sofer found a bug.
{title:References}
{p 4 8 2}Jones, M.C. 2004. Hazelton, M.L. (2003), "A graphical tool
for assessing normality," {it:The American Statistician}
57: 285-288: Comment. {it:The American Statistician} 58: 176-177.
{p 4 8 2}Jones, M.C. and F. Daly. 1995. Density probability plots.
{it:Communications in Statistics, Simulation and Computation}
24: 911-927.
{title:Also see}
{p 4 13 2}On-line: help for {help twoway}, {help diagplots},
{help gumbelfit} (if installed),
{help gammafit} (if installed),
{help betafit} (if installed)
{p_end}