{smcl}
{hline}
help for {hi:kwstat}{right:{browse "http://econ.chavezjuarez.com/vcheck.php?i=kwstat&v=1.0":Version 1.0}}
{hline}

{title:Kernel weighted sample statistics}


{p 4 17 2}{cmd:kwstat} {it:yvar xvar} [if] [in] [,
	{cmdab:b:w(}{it:real}{cmd:)}
	{cmdab:lp:olybw}
	{cmdab:k:ernel(}{it:str}{cmd:)}
	{cmdab:s:tats(}{it:str}{cmd:)}
	{cmdab:at}
	{cmdab:grid(}{it:int}{cmd:)}
	{cmdab:save}	
	{cmdab:pre:fix(}{it:str}{cmd:)}
	{cmdab:no:graph}
	{cmdab:graphtype(}{it:str}{cmd:)}
	{cmdab:grapho:ptions(}{it:str}{cmd:)}
		]

{title:Description}

{p 4 4 4} {cmd:kwstat} computes sample statistics of {it:yvar} in function of another variable {it:xvar}. The approach is inspired by the kernel regression (Nadaraya-Watson estimator) which computes 
the conditional mean of y in function of x. {cmd:kwstat} does the same but not only for the mean but also for the standard deviation, deciles, range, etc. Note that this procedure is an
ad-hoc method and should be used in an exploratory way to visualize the data. It is not rooted in a well defined statistical concept. 

{title:options}

{p 4 4 4} 	{cmdab:b:w(}{it:real}{cmd:)} allows you to define the bandwidth of the kernel used in the computation. If nothing is specified, 
kwstat uses h = 1.06 {sigma} n^(-1/5), which is an approximation of the optimal bandwidth for the Gaussian kernel when estimating the Nadaraya-Watson estimator. 
Hence, it is not necessarily the optimal bandwidth for the estimation of other statistics and/or other kernels. 
I strongly advise users to try several bandwidths and to compare them visually. See also the user manual for more details. 

{p 4 4 4} 	{cmdab:lp:olybw} is a second option to modify the optimal bandwidth with respect to the default value. 
By activating this option the optimal bandwidth proposed  by the command {cmd:lpoly} is used. Note, however, that this is only possible for the kernels 
implemented in {cmd:lpoly} (e.g. it's not possible for the logit kernel) and that these optimized bandwidths were designed for the kernel regression 
(estimation of the mean). They are therefore not necessarily optimal for the estimation of other statistics. I strongly advise users to try several bandwidths 
and to compare them visually. See also section the user manual for more details.

{p 4 4 4} 	{cmdab:k:ernel(}{it:str}{cmd:)} allows you to change the kernel type. The options are as follows: 

{col 15} {it:default} {col 28} The default epanechnikov kernel
{col 15} normal {col 28} The normal density kernel
{col 15} triangle {col 28} The triangle kernel
{col 15} beta {col 28} The beta kernel
{col 15} logit {col 28} The logit density kernel
{col 15} uniform {col 28} The uniform kernel
{col 15} cosine {col 28} The cosine kernel
{col 15} parzen {col 28} The Parzen kernel

{p 4 4 4} 	{cmdab:s:tats(}{it:str}{cmd:)} lets you specify the statistics you would like to compute. The default value is the mean, providing equivalent results to 
{cmd:lpoly}. All statistics supported by the command {cmd:tabstat} are supported: {cmd:mean, sd, min, max, range, kurtosis, skewness, semean, p10, p95} etc. 
(e.g. see {cmd:help tabstat} for the full list). Several statistics can be selected together, for instance {cmd:stats(mean sd p10)} returns the mean, the 
standard deviation and the first decile of {it:yvar} in function of {it:xvar}.

{p 4 4 4} 	{cmdab:at} by default the statistics are evaluated on an equally spaced grid at 100 points. Using the option {cmd:at} you can change this and let 
{cmd:kwstat} compute the statistics for each value of the variable {it:xvar}. Note, however, that his can become computationally heavy when many different values are 
present. You can also change the number of points on the equally spaced grid using the option {cmd:grid({it:int})}

{p 4 4 4} 	{cmdab:grid(}{it:int}{cmd:)}  allows you to change the number of points for which the statistics are computed. By default, 100 points on the equally 
spaced grid between the minimum and the maximum of {it:xvar} are used.

{p 4 4 4} 	{cmdab:save} allows you to store the variables produces by {cmd:kwstat}. This can be helpful for posterior use (e.g. to create your own graph). 
You can also use the option {cmd:prefix} to change the name of the variables. By {it:default} the variables are names {cmd:_kwstats_{it:stat}} 
where {cmd:{it:stat}} refers to the statistic chosen in the option {cmd:stats}. 	
 
{p 4 4 4} 	{cmdab:pre:fix(}{it:str}{cmd:)} allows you to change the prefix of the generated variable from the default value of {cmd:_kwstat_} 
to the prefix of your preference.

{p 4 4 4} 	{cmdab:no:graph} suppresses the output of the graph. Note that this makes only sense if you also use the option {cmd:save} to keep 
the computed variables for posterior use. Otherwise the command would not produce any output.

{p 4 4 4} 	{cmdab:graphtype(}{it:str}{cmd:)} lets you change the graph type. By default a {it:line} graph is provided. 
You can change it to {it:scatter} or {it:connected}.

{p 4 4 4} 	{cmdab:grapho:ptions(}{it:str}{cmd:)} allows you to provide a string containing options for the twoway graph. You can for instance 
change the legend or the label of the axis. See {cmd:help twoway} for more details

{title:Output}
{p 4 4 15} {cmd:kwstat} returns several values which can be accessed through {stata return list:return list}:

{col 10} r(bw): {col 25} the bandwidth
{col 10} r(kernel): {col 25} the kernel function used in the computation



{title:Example}
{p 0 4 0}First, we load a sample database containing several thousand individuals:{break}{stata sysuse nlsw88, clear} 

{p 0 4 0}Then we can compute several statistics for {it:wage} in function of {it:tenure}{break}
{stata kwstat wage tenure if tenure<22, lpoly stats(mean median sd)}{break}
{stata kwstat wage tenure if tenure<22, lpoly stats(skewness) graphoptions( ytitle("Skewness") legend(on))}

{title:Author}
{p 4 4 4}	For suggestions and questions, please contact the author: {break} 
Florian Chávez Juárez: {browse "mailto:florian@chavezjuarez.com?subject=Stata kernel:":florian@chavezjuarez.com}

{title:Update and user manual}
{p 4 4 4}	The easiest way to update the package is: {stata ssc install kwstat, replace:ssc install kwstat, replace}{break}
If you want to see if you have the newest version and what changed,  {browse "http://econ.chavezjuarez.com/vcheck.php?i=kwstat&v=1.0":please click here}.{break}
Or download the {browse "http://econ.chavezjuarez.com/download.php?id=17&h=667a99":User manual} from website of the author.