Basic options for parmhet and parmiv
options Description ------------------------------------------------------------------------- by(varlist) Variables specifying by-groups eform Estimates and confidence limits exponentiated float Numeric output variables of type float or less dfcombine(combination_rule) Rule for combining degrees of freedom ivweight(newvarname) Name of generated inverse-variance weight variable sweight(newvarname) Name of generated semi-weight variable sstderr(newvarname) Name of generated semi-weight-based standard error variable -------------------------------------------------------------------------
where combination_rule is
welch | constant
Description
These options are used in a similar way by parmhet and parmiv.
Options
by(varlist) specifies a list of existing by-variables in the input dataset. The heterogeneity-test statistics are computed within by-groups if by() is specified. If by() is not specified, then the test statistics are computed for the whole input dataset. The output dataset (or resultsset) created by parmhet has one observation per by-group, if by() is specified, and only one observation otherwise.
eform indicates that the input estimates are exponentiated, and that the input standard errors are multiplied by the exponentiated estimate. eform is used when the user chooses to calculate heterogeneity-test statistics for the logs of the input parameter estimates, as is usual when the input parameters are odds ratios, risk ratios, or geometric mean ratios. Otherwise, heterogeneity-test statistics are calculated for the input parameters themselves.
float specifies that the numeric output variables will be created as type float or below. If float is unset, then the numeric output variables are created as type double. Note that all generated variables are compressed as much as possible without loss of information, whether or not float is specified.
dfcombine(combination_rule) specifies a rule for combining the degrees of freedom of the input parameters to define the denominator degrees of freedom for the F-test statistic, if a degrees of freedom variable is specified by the user. If dfcombine(welch) is specified, then parmhet and parmiv use the formula of Welch (1951), popularized by Cochrane (1954), to calculate the F-statistic and its denominator degrees of freedom. If dfcombine(constant) is specified, then parmhet and parmiv check that the input degrees of freedom are constant (or constant within by-groups if by(varlist) is specified), and then sets the denominator degrees of freedom to the constant input degrees of freedom, and calculates the heterogeneity F-statistic by dividing the heterogeneity chi-squared statistic by the heterogeneity degrees of freedom. dfcombine() is set to welch by default, but is ignored if a degrees of freedom variable is not specified. The option dfcombine(constant) is useful if the input parameters are uncorrelated parameters belonging to the same model estimation with pooled degrees of freedom, such as group means estimated using the regress command with group membership indicators as X-variables, using the noconst option, and the user uses parmhet or parmiv to test for heterogeneity between groups. In these circumstances, using regress without the robust option and using dfcombine(constant) with parmhet or parmiv gives P-values equivalent to those of the equal-variance F-test.
ivweight(newvarname) specifies the name of an output variable, to be generated in the existing input dataset, containing inverse-variance weights for the corresponding parameter estimates. These inverse-variance weights can then be input as aweights to the metaparm, module of the parmest package, using the estimates, standard errors and degrees of freedom input to parmhet or parmiv, to output estimates, confidence intervals and P-values for summary parameters generated by a fixed-effect meta-analysis.
sweight(newvarname) specifies the name of an output variable, to be generated in the existing input dataset, containing semi-weights for the corresponding parameter estimates, as described in Cochrane (1954). These semi-weights can then be input as aweights to the metaparm module of the parmest package, using the estimates and degrees of freedom input to parmhet or parmiv together with the standard errors generated using the sstderr() option, to output estimates, confidence intervals and P-values for summary parameters generated by a DerSimonian-Laird randomly-variable-effect meta-analysis, as defined by DerSimonian and Laird (1986).
sstderr(newvarname) specifies the name of an output variable, to be generated in the existing input dataset, containing semi-weight-based standard errors for the corresponding parameter estimates. These standard errors are equal to the inverse square roots of the semi-weights generated by the sweight() option. If calculated, they can be input as the stderr() option to the metaparm module of the parmest package, using the estimates and degrees of freedom input to parmhet or parmiv, together with aweights generated by the sweight() option, to output estimates, confidence intervals and P-values for summary parameters generated by a DerSimonian-Laird randomly-variable-effect meta-analysis, as defined by DerSimonian and Laird (1986).
Author
Roger Newson, National Heart and Lung Institute, Imperial College London, UK. Email: r.newson@imperial.ac.uk
References
Cochrane, W. G. 1954. The combination of estimates from different experiments. Biometrics 10(1): 101-129.
DerSimonian, R. and Laird, N. 1986. Meta-analysis in clinical trials. Controlled Clinical Trials 7(3): 177-188.
Welch, B. L. 1951. On the comparison of several mean values: an alternative approach. Biometrika 36(3/4): 330-336.
Also see
Manual: [R] meta, [R] test On-line: help for parmhet, parmiv, parmhet_resultsset_opts, parmhet_hettest_opts, parmhet_resultsset help for test help for parmest, parmby, parmcip, metaparm, metan if installed