mlmethod.hlp

是一个经济学管理应用软件 很难找的 但是经济学学生又必须用到

{smcl}
{* 15feb2005}{...}
{cmd:help mlmethod}
{hline}

{title:Title}

{p2colset 4 14 16 2}{...}
{p2col :{hi:[R] ml} {hline 2}}User-written evaluator programs for use with ml methods lf, d0, d1, and d2{p_end}
{p2colreset}{...}


{title:Descripton}

{pstd}
The {helpb ml} command requires a program to be written
to evaluate the log-likelihood function.  It is referred to as the user-written
evaluator.  The {cmd:ml} support commands {cmd:mleval}, {cmd:mlsum},
{cmd:mlvecsum}, {cmd:mlmatsum}, and {cmd:mlmatbysum} are helpful in writing
d0, d1, and d2 evaluator programs; see {help mleval}.


{title:Method lf evaluators}

{p 8 14 2}{cmd:program define} {it:progname}{p_end}
{p 12 18 2}{cmd:version {ccl stata_version}}{p_end}
{p 12 18 2}{cmd:args lnf theta1 theta2} {it:...}{p_end}
{p 12 18 2}/* if you need to create any intermediate results: */{p_end}
{p 12 18 2}{cmd:tempvar tmp1 tmp2} {it:...}{p_end}
{p 12 18 2}{cmd:quietly gen double `tmp1' =} {it:...}{p_end}
	    {it:...}
{p 12 18 2}{cmd:quietly replace `lnf' =} {it:...}{p_end}
	{cmd:end}

    where

{p 8 24 2}{cmd:`lnf'} {space 5} variable to be filled in with
observation-by-observation values of ln l_j{p_end}
{p 8 24 2}{cmd:`theta1'} {space 2} variable containing evaluation of 1st
equation theta_1j = x_1j * b_1{p_end}
{p 8 24 2}{cmd:`theta2'} {space 2} variable containing evaluation of 2nd
equation theta_2j = x_2j * b_2{p_end}
	...


{title:Method d0 evaluators}

{p 8 14 2}{cmd:program define} {it:progname}{p_end}
{p 12 16 2}{cmd:version {ccl stata_version}}{p_end}
{p 12 16 2}{cmd:args todo b lnf}{p_end}
{p 12 16 2}{cmd:tempvar theta1 theta2} {it:...}{p_end}
{p 12 16 2}{cmd:mleval `theta1' = `b', eq(1)}{p_end}
{p 12 16 2}{cmd:mleval `theta2' = `b', eq(2)}    /* if there is a theta2 */{p_end}
	    {it:...}
{p 12 16 2}/* if you need to create any intermediate results: */{p_end}
{p 12 16 2}{cmd:tempvar tmp1 tmp2} {it:...}{p_end}
{p 12 16 2}{cmd:gen double `tmp1' =} {it:...}{p_end}
	    {it:...}
{p 12 16 2}{cmd:mlsum `lnf' =} {it:...}{p_end}
	{cmd:end}

    where

{p 8 22 2}{cmd:`todo'} {space 2} always contains 0 (may be ignored){p_end}
{p 8 22 2}{cmd:`b'} {space 5} full parameter row vector b=(b_1,b_2,...,b_E){p_end}
{p 8 22 2}{cmd:`lnf'} {space 3} scalar to be filled in with overall lnL{p_end}


{title:Method d1 evaluators}

{p 8 14 2}{cmd:program define} {it:progname}{p_end}
{p 12 18 2}{cmd:version {ccl stata_version}}{p_end}
{p 12 18 2}{cmd:args todo b lnf g} [{cmd:negH g1} [{cmd:g2} {it:...}]]{p_end}
{p 12 18 2}{cmd:tempvar theta1 theta2} {it:...}{p_end}
{p 12 18 2}{cmd:mleval `theta1' = `b', eq(1)}{p_end}
{p 12 18 2}{cmd:mleval `theta2' = `b', eq(2)}    /* if there is a theta2 */{p_end}
	    {it:...}
{p 12 18 2}/* if you need to create any intermediate results: */{p_end}
{p 12 18 2}{cmd:tempvar tmp1 tmp2} {it:...}{p_end}
{p 12 18 2}{cmd:gen double `tmp1' =} {it:...}{p_end}
	    {it:...}
{p 12 18 2}{cmd:mlsum `lnf' =} {it:...}{p_end}
{p 12 18 2}{cmd:if `todo'==0 | `lnf'==. {c -(} exit {c )-}}{p_end}
{p 12 18 2}{cmd:tempname d1 d2} {it:...}{p_end}
{p 12 18 2}{cmd:mlvecsum `lnf' `d1' =} formula for
dlnl_j/dtheta_1j{cmd:, eq(1)}{p_end}
{p 12 18 2}{cmd:mlvecsum `lnf' `d2' =} formula for
dlnl_j/dtheta_2j{cmd:, eq(2)}{p_end}
	    {it:...}
{p 12 18 2}{cmd:matrix `g' = (`d1',`d2',} {it:...} {cmd:)}{p_end}
	{cmd:end}

    where

{p 8 23 2}{cmd:`todo'} {space 3} contains 0 or 1;  0 ==> {cmd:`lnf'} to be
filled in; 1 ==> {cmd:`lnf'} and {cmd:`g'} to be filled in{p_end}
{p 8 23 2}{cmd:`b'} {space 6} full parameter row vector b=(b_1,b_2,...,b_E){p_end}
{p 8 23 2}{cmd:`lnf'} {space 4} scalar to be filled in with overall lnL{p_end}
{p 8 23 2}{cmd:`g'} {space 6} row vector to be filled in with overall g =
dlnL/db{p_end}
{p 8 23 2}{cmd:`negH'} {space 3} argument to be ignored{p_end}
{p 8 23 2}{cmd:`g1'} {space 5} variable optionally to be filled in with
dlnl_j/db_1{p_end}
{p 8 23 2}{cmd:`g2'} {space 5} variable optionally to be filled in with
dlnl_j/db_2{p_end}
	...


{title:Method d2 evaluators}

{p 8 14 2}{cmd:program define} {it:progname}{p_end}
{p 12 18 2}{cmd:version {ccl stata_version}}{p_end}
{p 12 18 2}{cmd:args todo b lnf g negH} [{cmd:g1} [{cmd:g2} {it:...}]]{p_end}
{p 12 18 2}{cmd:tempvar theta1 theta2} {it:...}{p_end}
{p 12 18 2}{cmd:mleval `theta1' = `b', eq(1)}{p_end}
{p 12 18 2}{cmd:mleval `theta2' = `b', eq(2)}    /* if there is a theta2 */{p_end}
	    {it:...}
{p 12 18 2}/* if you need to create any intermediate results: */{p_end}
{p 12 18 2}{cmd:tempvar tmp1 tmp2} {it:...}{p_end}
{p 12 18 2}{cmd:gen double `tmp1' =} {it:...}{p_end}
	    {it:...}
{p 12 18 2}{cmd:mlsum `lnf' =} {it:...}{p_end}
{p 12 18 2}{cmd:if `todo'==0 | `lnf'==. {c -(} exit {c )-}}{p_end}
{p 12 18 2}{cmd:tempname d1 d2} {it:...}{p_end}
{p 12 18 2}{cmd:mlvecsum `lnf' `d1' =} formula for
dlnl_j/dtheta_1j{cmd:, eq(1)}{p_end}
{p 12 18 2}{cmd:mlvecsum `lnf' `d2' =} formula for
dlnl_j/dtheta_2j{cmd:, eq(2)}{p_end}
	    {it:...}
{p 12 18 2}{cmd:matrix `g' = (`d1',`d2',} {it:...} {cmd:)}{p_end}
{p 12 18 2}{cmd:if `todo'==1 | `lnf'==. {c -(} exit {c )-}}{p_end}
{p 12 18 2}{cmd:tempname d11 d12 d22} {it:...}{p_end}
{p 12 18 2}{cmd:mlmatsum `lnf' `d11' =} formula for
-d^2lnl_j/dtheta_1j^2{cmd:, eq(1)}{p_end}
{p 12 18 2}{cmd:mlmatsum `lnf' `d12' =} formula for -d^2lnl_j/(dtheta_1j
dtheta_2j){cmd:, eq(1,2)}{p_end}
{p 12 18 2}{cmd:mlmatsum `lnf' `d22' =} formula for
-d^2lnl_j/dtheta_2j^2{cmd:, eq(2)}{p_end}
	    {it:...}
{p 12 18 2}{cmd:matrix `negH' = (`d11',`d12',} {it:...} {cmd:\ `d12'',`d22',}
{it:...} {cmd:)}{p_end}
	{cmd:end}

    where

{p 8 22 2}{cmd:`todo'} {space 2} contains 0, 1, or 2;  0 ==> {cmd:`lnf'} to be
filled in;{p_end}
{p 22 22 2}1 ==> {cmd:`lnf'} and {cmd:`g'} to be filled in;{p_end}
{p 22 22 2}2 ==> {cmd:`lnf'}, {cmd:`g'}, {cmd:`negH'} to be filled in{p_end}
{p 8 22 2}{cmd:`b'} {space 5} full parameter row vector b=(b_1,b_2,...,b_E){p_end}
{p 8 22 2}{cmd:`lnf'} {space 3} scalar to be filled in with overall lnL{p_end}
{p 8 22 2}{cmd:`g'} {space 5} row vector to be filled in with overall g =
dlnL/db{p_end}
{p 8 22 2}{cmd:`negH'} {space 2} matrix to be filled in with overall negative
Hessian -H = -d^2lnL/dbdb'{p_end}
{p 8 22 2}{cmd:`g1'} {space 4} variable optionally to be filled in with
dlnl_j/db_1{p_end}
{p 8 22 2}{cmd:`g2'} {space 4} variable optionally to be filled in with
dlnl_j/db_2{p_end}
	...


{title:Global macros for use by all evaluators}

{p 8 23 2}{cmd:$ML_y1}{space 5}name of first dependent variable{p_end}
{p 8 23 2}{cmd:$ML_y2}{space 5}name of second dependent variable, if any{p_end}
	...
{p 8 23 2}{cmd:$ML_samp}{space 3}variable containing 1 if observation to be used;
0 otherwise{p_end}
{p 8 23 2}{cmd:$ML_w}{space 6}variable containing weight associated with
observation or 1 if no weights specified{p_end}

{pstd}
Method lf evaluators can ignore {cmd:$ML_samp}, but restricting
calculations to the {cmd:$ML_samp==1} subsample will speed execution.  Method
lf evaluators must ignore {cmd:$ML_w}; application of weights is handled
by the method itself.

{pstd}
Method d0, d1, and d2 can ignore {cmd:$ML_samp} as long as
{cmd:ml model}'s {cmd:nopreserve} option is not specified.  Method d0,
d1, and d2 will run more quickly if {cmd:nopreserve} is specified.
Method d0, d1, and d2 evaluators can ignore {cmd:$ML_w} only
if they use {cmd:mlsum}, {cmd:mlvecsum}, and {cmd:mlmatsum} to produce final
results; see {help mleval}.


{title:Also see}

{psee}
Manual:  {bf:[R] ml}

{psee}
Online:  {helpb ml}, {help mleval}; {helpb macro},
{helpb matrix}, {helpb program}
{p_end}