secoefs([Term] [, errorTerm:ErrorTerm, byterm:F, se:F or coefs:F, silent:T]), Term a CHARACTER scalar, a positive integer, or a factor or variate in the current GLM model, ErrorTerm a CHARACTER scalar or positive integer. byterm:F only when Term, se:F and coefs:F omitted |

secoefs(Term) returns the model effects or regression coefficients and their standard errors for the term specified in the CHARACTER variable Term. The result is a structure with components 'coefs' and 'se'. The coefficients and standard errors pertain to the results of the most recent GLM (generalized linear or linear model) command such as regress(), anova(), or poisson(). When Term is a main effect term, the components are vectors. When it is an interation term, the components are matrices or arrays with the leftmost subscript corresponding to the leftmost factor in Term. Caution: After anova(), manova() and regress(), standard errors are computed using the final error mean square in the model. This may not be appropriate with mixed models, including split plot designs. Term is usually a quoted string or CHARACTER variable such as "a.b" which exactly matches a term in the most recent model, that is, "a.b" is not the same as "b.a". An interaction term produces a matrix or array with the leftmost subscript corresponding to the leftmost factor in Term. If any variables in Term originally specified in the form {expr}, where expr is a MacAnova expression, you must include the enclosing '{' and '}'. For a term which consists of a single factor or variate, Term can be its unquoted name. Alternatively, Term can be a integer between 1 and the number of terms, excluding the final error term. For example, unless the model contained "-1", secoefs(1) gets the estimated intercept or grand mean and its standard error. secoefs() (no Term specified) computes coefficients and standard errors for all terms in the model. The result is a structure with one component for each term in the model, with each component itself a structure with components 'coefs' and 'se'. The names of the top level components in the result are taken from the names of the terms, truncated if necessary to 12 characters. When such truncation is neccessary, the result is also given labels which contain the full component names. See topic 'labels'. secoefs(byterm:F) is the same as secoefs() except that the resulting structure has two components, 'coefs' and 'se', each of which is a structure with one component per term (unless there is only one term in the model). In this case, the names of the bottom level components are taken from the possibly truncated names of the terms. When any truncation takes place, the full term names are also attached as labels. secoefs(Term, silent:T) and secoefs(silent:T) do the same, but certain warning and advisory messages are suppressed. 'silent:T' can be used with any other keywords. This feature is useful in a macro when warning messages might confuse the user, or in a simulation. The default value of 'silent' is False unless the value of option' 'warnings' is False. secoefs(Term,coefs:F) and secoefs(coefs:F) (or secoefs(,coefs:F)) suppress the computation of the coefficients, returning a structure or matrix containing only standard errors. secoefs(Term,se:F) and secoefs(se:F) are equivalent to coefs(Term) and coefs(), respectively. You can compute a structure containint t-statistics for every coefficient by Cmd> tt <- secoefs(se:F)/secoefs(coefs:F) #or coefs()/secoefs(coefs:F) Alternatively, you could compute such a structure by Cmd> @tmp <- secoefs(byterm:F); tt <- @tmp$coefs/@tmp$se secoefs(Term,Varno) or secoefs(,Varno) computes coefficients and standard errors only for variable number Varno in the case of a multivariate dependent variable. If present, Varno must be the second argument and any keywords must follow it. For all forms, an optional keyword phrase argument errorterm:ErrTerm or errorterm:ErrTermNo, where ErrTerm is a CHARACTER variable or quoted string specifying a term in the model and ErrTermNo is a positive integer, specifies that the MS from the indicated term is to be used in computing standard errors. If tcrit is a critical value, say, invstu(1-alpha/2,errorDF), you can compute the lower 1-alpha confidence limits for the coefficients Cmd> @tmp <- secoefs(byterm:F);@tmp$coefs - tcrit*@tmp$se and similarly for upper limits (replace - by +). Example: After anova("y= a + b + a.b") secoefs(a), secoefs("a"), or secoefs(1) will compute the main effect coefficients and their standard errors for factor a secoefs(a,coefs:F), secoefs("a",coefs:F), or secoefs(2,coefs:F) will compute the standard errors of main effect coefficients secoefs("a.b") or secoefs(4) will produce matrices of the a by b interaction coefficients and their standard errors. secoefs() will produce all coefficients and their standard errors in a structure with components CONSTANT, a, b, and a.b. secoefs(byterm:F) will produce all coefficients and their standard errors in a structure with components coefs and se This will produce the a by b interaction effects and their standard errors. Secoefs() does not work after fastanova(), ipf(), or screen(), or if 'coefs:F' was an argument to the most recent GLM command. See also coefs(), contrast(), modelinfo(), popmodel(), pushmodel().

Gary Oehlert 2003-01-15