{smcl} {* *! version 1.1.9 19 September 2008}{...} {cmd:help qic} {right:also see: {help xtgee }} {hline} {title:Title} {p2colset 5 19 21 2}{...} {p2col :{hi: qic} {hline 2}}QIC criterion for model selection in GEE analyses{p_end} {p2colreset}{...} {title:Syntax} {p 8 14 2} {cmd:qic} {depvar} [{indepvars}] {ifin} [{cmd:,} {it:options}] {synoptset 21 tabbed}{...} {synopthdr} {synoptline} {syntab:Model} {synopt :{opth "i(varname:varname_i)"}}use {it:varname_i} as the panel ID variable{p_end} {synopt :{opth "t(varname:varname_t)"}}use {it:varname_t} as the time variable{p_end} {synopt :{cmdab:f:amily(}{it:{help qic##family:family}}{cmd:)}}distribution of {depvar}{p_end} {synopt :{cmdab:l:ink(}{it:{help qic##link:link}}{cmd:)}}link function{p_end} {syntab:Model 2} {synopt :{opth e:xposure(varname)}}include ln({it:varname}) in model with coefficient constrained to 1{p_end} {synopt :{opth off:set(varname)}}include {it:varname} in model with coefficient constrained to 1{p_end} {synopt :{opt nocon:stant}}suppress constant term{p_end} {synopt :{opt force}}estimate even if observations unequally spaced in time{p_end} {syntab:Correlation} {synopt :{cmdab:c:orr(}{it:{help qic##correlation:correlation}}{cmd:)}}within-group correlation structure{p_end} {syntab:SE/Robust} {synopt :{opt r:obust}}synonym for {cmd:vce(robust)}{p_end} {synopt :{opt nmp}}use divisor N-P instead of the default N{p_end} {synopt :{opt rgf}}multiply the robust variance estimate by (N-1)/(N-P){p_end} {synopt :{cmdab:s:cale(x2)}}set scale parameter to Pearson chi-squared statistic{p_end} {synopt :{cmdab:s:cale(dev)}}set scale parameter to deviance divided by degrees of freedom{p_end} {synopt :{cmdab:s:cale(phi)}}do not rescale the variance{p_end} {synopt :{opt scale(#)}}set scale parameter to {it:#}{p_end} {syntab:Reporting} {synopt :{opt level(#)}}set confidence level; default is {cmd:level(95)}{p_end} {synopt :{opt ef:orm}}report exponentiated coefficients{p_end} {syntab:Opt options} {synopt :{it:{help qic##optimize_options:optimize_options}}}control the optimization process; seldom used{p_end} {p2coldent :{opt nodis:play}}suppress display of header and coefficients{p_end} {synoptline} {p2colreset}{...} {p 4 6 2}{it:depvar} and {it:indepvars} may contain time-series operators; see {help tsvarlist}.{p_end} {p 4 6 2} {opt iweight}s, {opt fweight}s, and {opt pweight}s are allowed; see {help weight}. Weights must be constant within panel.{p_end} {marker family}{...} {synoptset 23}{...} {synopthdr :family} {synoptline} {synopt :{opt gau:ssian}}Gaussian (normal); {cmd:family({opt nor:mal})} is a synonym{p_end} {synopt :{opt ig:aussian}}inverse Gaussian{p_end} {synopt :{opt b:inomial}}Bernoulli/binomial (k=1){p_end} {synopt :{opt po:isson}}Poisson{p_end} {synopt :{opt nb:inomial}}negative binomial (k=1){p_end} {synopt :{opt gam:ma}}gamma{p_end} {synoptline} {p2colreset}{...} {marker link}{...} {synoptset 23}{...} {synopthdr :link} {synoptline} {synopt :{opt i:dentity}}identity; y=y{p_end} {synopt :{opt log}}log; ln(y){p_end} {synopt :{opt logi:t}}logit; ln{y/(1-y)}, natural log of the odds{p_end} {synopt :{opt p:robit}}probit; inverse Gaussian cumulative{p_end} {synopt :{opt cl:oglog}}cloglog; ln{-ln(1-y)}{p_end} {synopt :{opt pow:er}[{it:#}]}power; y^k with k=#; #=1 if not specified{p_end} {synopt :{opt opo:wer}[{it:#}]}odds power; [{y/(1-y)}^k - 1]/k with k=#; #=1 if not specified{p_end} {synopt :{opt nb:inomial}}negative binomial{p_end} {synopt :{opt rec:iprocal}}reciprocal; 1/y{p_end} {synoptline} {p2colreset}{...} {marker correlation}{...} {synoptset 23}{...} {synopthdr :correlation} {synoptline} {synopt :{opt exc:hangeable}}exchangeable{p_end} {synopt :{opt ind:ependent}}independent{p_end} {synopt :{opt uns:tructured}}unstructured{p_end} {synopt :{opt fix:ed} {it:matname}}user-specified{p_end} {synopt :{opt ar} {it:#}}autoregressive of order {it:#}{p_end} {synopt :{opt sta:tionary} {it:#}}stationary of order {it:#}{p_end} {synopt :{opt non:stationary} {it:#}}nonstationary of order {it:#}{p_end} {synoptline} {p2colreset}{...} {title:Description} {pstd} {cmd:qic} calculates the QIC and QIC_u criteria for model selection in GEE, which is an extension of the widely used AIC criterion in ordinary regression (Pan 2001). It allows for specification of all 7 distributions - gaussian, inverse Gaussian, Bernoulli/binomial, Poisson, negative binomial and gamma, all link functions and working correlation structures and all se/robust options, except for the vce option, avaiable in Stata 9.0. It also calculates the trace of the matrix O^{-1}V, where O is the variance estimate under the independent correlation structure and V is the variance estimate under the specified working correlation structure in GEE. When trace is close to the number of parametr p, the QIC_u is a good approximation to QIC. {title:Options} {dlgtab:Model} {phang} {opth "i(varname:varname_i)"}, {opth "t(varname:varname_t)"}; see {help estimation options##i():estimation options}. {pmore} {cmd:qic} does not need to know {opt t()} for the {cmd:corr(independent)} and {cmd:corr(exchangeable)} correlation structures. Whether you specify {opt t()} makes no difference in these two cases. {phang} {opt family(family)} specifies the distribution of {depvar}; {cmd:family(gaussian)} is the default. {phang} {opt link(link)} specifies the link function; the default is the canonical link for the {opt family()} specified. {dlgtab:Model 2} {phang} {opth exposure(varname)} and {opth offset(varname)} are different ways of specifying the same thing. {opt exposure()} specifies a variable that reflects the amount of exposure over which the {depvar} events were observed for each observation; ln({it:varname}) with coefficient constrained to be 1 is entered into the regression equation. {opt offset()} specifies a variable that is to be entered directly into the log-link function with its coefficient constrained to be 1; thus, exposure is assumed to be e^varname. If you were fitting a Poisson regression model, {cmd:family(poisson) link(log)}, for instance, you would account for exposure time for specifying {opt offset()} containing the log of exposure time. {phang} {opt noconstant} specifies that the linear predictor has no intercept term, thus forcing it through the origin on the scale defined by the link function. {phang} {opt force} specifies that estimation be forced even though {opt t()} is not equally spaced. This is relevant only for correlation structures that require knowledge of {opt t()} and that require observations be equally spaced. {dlgtab:Correlation} {phang} {opt corr(correlation)}; see {help estimation options##corr():estimation options}. {dlgtab:SE/Robust} {phang} {opt robust} specifies that the Huber/White/sandwich estimator of variance is to be used in place of the default GLS variance estimator; This produces valid standard errors even if the correlations within group are not as hypothesized by the specified correlation structure. It does, however, require that the model correctly specifies the mean. As such, the resulting standard errors are labeled "semi-robust" instead of "robust". Note that although there is no {opt cluster()} option, results are as if there were a {opt cluster()} option and you specified clustering on {opt i()}. {phang} {opt nmp}; see {help estimation options##nmp:estimation options}. {phang} {opt rgf} specifies that the robust variance estimate is multiplied by (N-1)/(N-P), where N = # of observations, and P = # of coefficients estimated. This option can be used only with {cmd:family(gaussian)} when {opt robust} is either specified or implied by the use of {opt pweight}s. Using this option implies that the robust variance estimate is not invariant to the scale of any weights used. {phang} {cmd:scale(x2}|{cmd:dev}|{it:#}|{cmd:phi)} overrides the default scale parameter of {cmd:scale(1)}; see {help estimation options##scale():estimation options}. {dlgtab:Reporting} {phang} {opt level(#)}; see {help estimation options##level():estimation options}. {phang} {opt eform} displays the exponentiated coefficients and corresponding standard erros and confidence intervals as described in {helpb maximize}. For {cmd:family(binomial) link(logit)} (i.e., logistic regression), exponentiation results in odds ratios; for {cmd:family(poisson) link(log)} (i.e., Poisson regression), exponentiated coefficients are incidence-rate ratios. {dlgtab:Opt options} {phang} {marker optimize_options} {it:optimize_options} control the iterative optimization process. These options are seldom used. {pmore} {opt iter:ate(#)} specifies the maximum number of iterations. When the number of iterations equals #, the optimization stops and presents the current results, even if the convergence tolerance has not been reached. The default value of {opt iterate()} is 100. {pmore} {opt tol:erance(#)} specifies the tolerance for the coefficient vector. When the relative change in the coefficient vector from one iteration to the next is less than or equal to #, the optimization process is stopped. {cmd:tolerance(1e-6)} is the default. {pmore} {opt nolog} suppress the display of the iteration log. {pmore} {opt tr:ace} specifies that the current estimates should be printed at each iteration. {phang} {opt nodisplay} suppresses the display of the header and coefficients. {title:Examples1} {phang}{stata "use http://www.stata-press.com/data/r9/nlswork2, clear":use http://www.stata-press.com/data/r9/nlswork2, clear} {phang}{stata iis id} {phang}{stata qic ln_w grade age if race == 2} {phang}{stata qic ln_w grade age, t(year) corr(uns) scale(dev) force nolog nodis trace} {phang}{stata qic ln_w grade age, t(year) corr(exc) force} {title:Examples2} {phang}{stata "use http://www.stata-press.com/data/r9/union, clear":use http://www.stata-press.com/data/r9/union, clear} {phang}{stata iis idcode} {phang}{stata tis year} {phang}{stata qic union age grade not_smsa south if black == 1, fam(bin)} {phang}{stata qic union age grade not_smsa south, fam(bin) link(probit) corr(uns) force tol(1e-8) iter(20)} {phang}{stata qic union age grade not_smsa south, fam(bin) link(cloglog) corr(ar) force scale(x2)} {title:Examples3} {phang}{stata "use http://www.stata-press.com/data/r9/ships, clear":use http://www.stata-press.com/data/r9/ships, clear} {phang}{stata egen wave = group(yr_con yr_op)} {phang}{stata iis ship} {phang}{stata tis wave} {phang}{stata qic accident op_75_79 co_65_69 co_70_74 co_75_79 if wave <= 6, fam(poi) corr(exc) ex(service)} {phang}{stata qic accident op_75_79 co_65_69 co_70_74 co_75_79, fam(poi) corr(sta) ex(service) force tol(1e-10) scale(dev)} {phang}{stata qic accident op_75_79 co_65_69 co_70_74 co_75_79, fam(poi) corr(exc) ex(service) force nodis} {title:Examples4} {phang}{stata "use http://www.stata-press.com/data/r9/airacc, clear":use http://www.stata-press.com/data/r9/airacc, clear} {phang}{stata iis airline} {phang}{stata tis time} {phang}{stata qic i_cnt inprog if airline <= 15, fam(nb 2) corr(exc) exposure(pmiles)} {phang}{stata qic i_cnt inprog, fam(nb 2) corr(sta) exposure(pmiles) force tol(1e-8) nodis} {phang}{stata qic i_cnt inprog, fam(nb 2) corr(uns) exposure(pmiles) force scale(x2)} {phang}{stata qic i_cnt inprog, fam(gam) corr(sta) exposure(pmiles) force scale(dev)} {phang}{stata qic i_cnt inprog, fam(ig) corr(uns) exposure(pmiles) force} {title:Also see} {psee} Manual: {bf:[XT] xtgee} {psee} Online: {help xtgee postestimation};{break} {helpb glm}, {helpb logistic}, {helpb prais}, {helpb regress}, {helpb svy}, {help xt}, {helpb xtcloglog}, {helpb xtdata}, {helpb xtdes}, {helpb xtgls}, {helpb xtintreg}, {helpb xtlogit}, {helpb xtnbreg}, {helpb xtpcse}, {helpb xtpoisson}, {helpb xtprobit}, {helpb xtreg}, {helpb xtregar}, {helpb xtsum}, {helpb xttab}, {helpb xttobit} {p_end} {title:Reference} {phang} Cui J. QIC program and model selection in GEE analyses. {it: Stata Journal} 2007; 7:209-220.{p_end} {phang} Cui J and Qian G. Selection of working correlation structure and best model in GEE analyses of longitudinal data. {it: Communications in Statistics, Simulation and Computation} 2007; 36:987-996.{p_end} {phang} Cui J and Feng L. Correlation structure and model selection for negative binomial distribution in GEE. {it: Communications in Statistics, Simulation and Computation} 2008 (in press).{p_end} {phang} Pan W. Akaike's information criterion in generalized estimating equations. {it:Biometrics} 2001; 57:120-125.{p_end} {title:Author} {p 4 4 2} James Cui, WHO Collaborating Centre for Obesity Prevention, Deakin University. {p 4 4 2} Email: {browse "mailto:jisheng.cui@deakin.edu.au":jisheng.cui@deakin.edu.au} Other Commands I have written: {help genhwcci} (if installed) {stata ssc install genhwcci} (to install this command) {help simuped2} (if installed) {stata ssc install simuped2} (to install this command) {help simuped3} (if installed) {stata ssc install simuped3} (to install this command) {help phenotype} (if installed) {stata ssc install phenotype} (to install this command) {help buckley} (if installed) {stata ssc install buckley} (to install this command)