{smcl}
{* 10sep2001/3nov2004}{...}
{hline}
help for {hi:anovaplot}
{hline}

{title:Plots of ANOVA fit, including interaction plots} 

{p 8 17 2}
{cmd:anovaplot} 
[{cmd:,} 
{cmd:scatter(}{it:scatter_options}{cmd:)} 
{it:connected_options} 
{cmd:plot(}{it:plot}{cmd:)} ] 
]   

{p 8 17 2}
{cmd:anovaplot} 
{it:varlist} 
[{cmd:,} 
{cmd:scatter(}{it:scatter_options}{cmd:)} 
{it:connected_options} 
{cmd:plot(}{it:plot}{cmd:)} ] 
]    


{title:Description}

{p 4 4 2}{cmd:anovaplot} plots fitted or predicted values from an immediately
previous one-, two- or three-way {cmd:anova}. By default the data for the
response are also plotted. In particular, {cmd:anovaplot} can show interaction
plots. 


{title:Remarks} 

{p 4 4 2}With the first syntax, no {it:varlist} is specified.  {cmd:anovaplot}
shows the response and predicted values on the {it:y} axis; the predictor named
first in the {cmd:anova} command on the {it:x} axis; any predictor named second
in the {cmd:anova} command by separate groups within a twoway plot;
and any predictor named third in the {cmd:anova} command by separate images
using a {cmd:by()} option. Thus with this syntax the appearance of the plot,
although not the values shown, is sensitive to the order in which predictors
are specified in the {cmd:anova} command. 

{p 4 4 2}With the second syntax, a {it:varlist} is supplied, which must be a
permutation of the predictors previously used. The order of the predictors in
the varlist is used to determine the form of the graph, as above. This permits
changing graphs without reissuing the {cmd:anova} command. 

{p 4 4 2}Note especially that the graph format produced by {cmd:anovaplot} is
appropriate for models with at most one continuous predictor, which should
always be the predictor named first. With that caveat, {cmd:anovaplot} offers a
way of showing parallel and diverging regression lines for models with one
continuous predictor.


{title:Options}

{p 4 8 2}
{cmd:scatter()} specifies options of {help scatter} which should be
specified to control plotting of fitted values. 

{p 8 8 2} 
To suppress plotting of response data, specify {cmd:scatter(ms(i))}.

{p 8 8 2} 
To separate tied values of response, specify {cmd:scatter(jitter(1))}. 

{p 4 8 2} 
{it:connected_options} are options of {help twoway connected} which 
should be specified to control plotting of fitted values. 

{p 4 8 2}{cmd:plot(}{help plot_option:plot}{cmd:)} provides a way to add other
plots to the generated graph; see {help plot_option}. 


{title:Examples}

{p 4 8 2}{cmd:. webuse systolic} 

{p 4 4 2}One-way anova: {cmd:systolic} as response on {it:y} axis, {cmd:drug} as
factor on {it:x} axis:

{p 4 8 2}{cmd:. anova systolic drug}{p_end}
{p 4 8 2}{cmd:. anovaplot}

{p 4 4 2}
Two-way anova, no interaction term: {cmd:systolic} as response on {it:y} axis, 
{cmd:disease} and {cmd:drug} as factors, with {cmd:disease} on {it:x} axis, 
and levels of {cmd:drug} defining separate profiles: 

{p 4 8 2}{cmd:. anova systolic disease drug}{p_end}
{p 4 8 2}{cmd:. anovaplot}

{p 4 4 2}Better to reverse order of predictors: 

{p 4 8 2}{cmd:. anovaplot drug disease}

{p 4 4 2}Example of interaction term: 

{p 4 8 2}{cmd:. anova systolic drug disease drug * disease}{p_end}
{p 4 8 2}{cmd:. anovaplot}

{p 4 4 2}
As above, but no plotting of data: 

{p 4 8 2}{cmd:. anovaplot, scatter(ms(i))}

{p 4 4 2}Examples with one continuous predictor: 

{p 4 8 2}{cmd:. sysuse auto}{p_end}
{p 4 8 2}{cmd:. decode foreign, gen(Foreign)}{p_end}
{p 4 8 2}{cmd:. gen F = substr(Foreign,1,1)}

{p 4 4 2}The continuous predictor is named just after the response: 

{p 4 8 2}{cmd:. anova price weight foreign, cont(weight)} 

{p 4 4 2}(N.B. parallel regression lines) 

{p 4 8 2}{cmd:. anovaplot, scatter(ms(i) mla(F) mlabpos(0))}

{p 4 8 2}{cmd:. anova price weight foreign weight*foreign, cont(weight)}

{p 4 4 2}(N.B. diverging regression lines) 

{p 4 8 2}{cmd:. anovaplot, scatter(ms(i) mla(F) mlabpos(0))}


{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}Ken Higbee provided very helpful comments. 


{title:Also see}

{p 4 13 2}On-line:  help for 
{help anova}, {help dotplot}, {help modeldiag}