{smcl}
{* $Date: 2008-08-20 $}{...}
{hline}
help for {hi:jonter}{right:Version 2.4 2008-08-20}
{hline}
{title:Jonckheere-Terpstra Test}
{p 4 11} {cmd:jonter}
{it:varname}
[{cmd:if} {it:exp}]
[{cmd:in} {it:exp}]
{cmd:,}
{cmd:by(}{it:groupvar}{cmd:)} [{cmdab:v:erbose} {cmdab:j:only} {cmdab:c:ontinuity}]
{title:Description}
{p}
{cmd:jonter} calculates test statistics, {it:J} and {it:J*}, and
asymptotic {it:p}-values for the Jonckheere-Terpstra test for ordered
alternatives.
{title:Options}
{p}
Note that the {cmd:by()} is mandatory. Both {it:varname}
and {it:groupvar} must be at least ordinal.
{p}
{cmd:verbose} allows display of the individual
Mann-Whitney U counts.
{p}
{cmd:jonly} limits the calculations to those for the {it:J} test statistic. This
allows the command to complete somewhat faster, and is highly recommended
for use of {cmd:jonter} with {cmd:permute} when use is made only of the {it:J} test statistic.
{p}
{cmd:continuity} causes a continuity correction to be applied to {it:J*}. This
might aid somewhat in the normal approximation, but use of a continuity correction
with the Jonckheere-Terpstra test is not widespread.
{title:Remarks}
{p}
{cmd:jonter} provides an alternative to the nonparametric trend test
{cmd:nptrend}, which is Stata's implementation of the Cuzick
test. The Jonckheere-Terpstra test's {it:p}-value asymptotically converges to
that for the test for Kendall's tau ({cmd:ktau}). As with {cmd:ktau}, {cmd:jonter} is
intended for small to moderate-sized datasets; the duration of computations grows
disproportionately with sample size.
{p}
{cmd:jonter} corrects {it:J} and {it:J*} for tied values in {it:varname}. It also reports
asymptotic {it:p}-values for the test of descending ordered alternatives, as well as for
that of ascending ordered alternatives. This obviates the need to manually reverse
ordering of the values in the {it:by} or {it:varname} variables in order to test for a
descending ordering. It also reports the asymptotic {it:p}-value for the alternative of
either an ascending or a descending ordering against the null hypothesis of no ordering
(i.e., the two-tailed test), when the direction of ordering in the {it:by} variable cannot
be independently specified in advance.
{p}
Test statistics and {it:p}-values are returned in scalars:
r(J){c 151}Jonckheere-Terpstra test statistic, {it:J};
r(JH0){c 151}expectation of {it:J} under the null hypothesis;
r(se_JH0){c 151}square-root of the variance of {it:J} under the null hypothesis;
r(Jstar){c 151}asymptotic normal approximation of {it:J}; and
r(p), r(p_l) and r(p_u){c 151}{it:p}-values associated with {it:J*}.
{title:Examples}
{p 4}{inp:. jonter latency, by(dose)}
{title:Author}
Joseph Coveney
jcoveney@bigplanet.com
{title:Acknowledgment}
{p}
Nicholas J. Cox edited an earlier version of {cmd:jonter}, fixing bugs
and providing major improvements in efficiency and functionality. Correspondence
regarding support should nevertheless be addressed to the author listed above.
{title:Also see}
Manual: {hi:[R] nptrend} {hi:[R] spearman}
{p 0 19}On-line: help for {help nptrend}, {help ktau}, {help somersd} (if installed)