Contents

estimlimits()

Usage:

estimlimits(x, confLevel), REAL vector or matrix x with no MISSING
  elements, 0 < confLevel < 1 a REAL scalar
estimlimits(NULL,factorValues, confLevel), REAL vector or matrix
  factorValues with no MISSING elements
estimalimits(x,factorValues, confLevel)

Keywords: regression, confidence limits

Macro estimlimits() computes confidence limits for the expected value
E(y|x) after a regression.  You can also use it to compute limits for
the expectation of y after anova() when the model includes one or more
factors.

Before using estimlimits(), you must have run regress("y=x") or
regress("y=x1 + x2 + ...  xk") or more generally anova(Model) where
Model may contain both one or more predictors (covariates) and/or one
or more factors

When the arguments to estimlimits() specify a single condition under
which limits are wanted for E(y), the result is vector(lower,upper),
where lower and upper are the limits.

When the arguments specify several conditions (several values of x
and/or several factor levels), the result is hconcat(lower,upper),
where lower and upper are vectors with length(lower) = length(upper) =
number of conditions.

                      Usage after regress()
estimlimits(x, confLevel), where REAL variable x is a scalar (simple
linear regression) or a vector (multiple regression) returns
vector(lower, upper), the confidence limits for E(y|x) for the
specified value of of the predictor variable(s).  x can contain no
MISSING elements and confLevel must be a real scalar between 0 and 1.

When there is more than one predictor variable, length(x) = number of
predictors.

When there is only one predictor (simple linear regression), x can be a
vector.  When there is more than one predictor, x can be a matrix with
ncols(x) = number of predictors.  In both cases, the result is
hconcat(lower,upper).

When confLevel < .5, a warning message is printed and the confidence
level is assumed to be 1 - confLevel.

            Usage after anova() with factors in the model
estimlimits(NULL,FactorValues,confLevel) is appropriate after anova()
with a model with no variates (covariates).

When there is just one factor ("y = a"), FactorValues should be a
scalar or vector of permissible factor levels.  When there are nFactors
> 1 factors, FactorValues should be a vector with length(FactorValues) =
nFactors, or a matrix with ncols(FactorValues) = nFactors.

estimlimits(x,FactorValues,confLevel) is appropriate after anova() with
a model containing both factors and (co)variates.  x and FactorValues
are as in the no factor case and no variate case, respectively.

Examples:
After regress("y=x1 + x2 + x3"):
 estimlimits(vector(2,3,4),.95) returns where lower and upper are the
 limits when x1 = 2, x2 = 3 and x3 = 4.

 estimlimits(hconcat(x01, x02, x03),.95), returns hconcat(lower,upper),
 where lower[I] and upper[I] are limits whenx1 = x01[I], x2 = x02[I]
 and x3 = x03[I]. x01, x02, and x03 must all be vectors of the same
 length.

After anova("y = a + b")
  estimlimits(NULL,vector(1,2),.95) returns vector(lower,upper), where
  lower and upper are limits when a = 1 and b = 2

 After anova("y = x + a + b")
  estimlimits(vector(3,4),vconcat(vector(1,2)',vector(1,3)'),.95)
  returns hconcat(lower,upper), where lower[1] and upper[1] are limits
  when x = 3, a = 1 and b = 2 and lower[2] and upper[2] are limits
  when x = 4, a= 1 and b = 3.

See also predlimits(), regpred(), glmpred().