clarify (lack of) centering when there is no intercept

Author: bgoodri

DESCRIPTION

@@ -1,8 +1,8 @@
 Package: rstanarm
 Type: Package
 Title: Bayesian Applied Regression Modeling via Stan
-Version: 2.14.2
-Date: 2017-04-12
+Version: 2.14.3
+Date: 2017-04-13
 Authors@R: c(person("Jonah", "Gabry", email = "[email protected]", role = "aut"),
              person("Imad", "Ali", role = "ctb"),
              person("Trustees of", "Columbia University", role = "cph"),

R/stan_biglm.R

@@ -23,9 +23,7 @@
 #' 
 #' @export
 #' @param biglm The list output by \code{\link[biglm]{biglm}} in the \pkg{biglm}
-#'   package. The original call to \code{\link[biglm]{biglm}} must not have an 
-#'   intercept and must utilize \emph{centered} but not \emph{standardized} 
-#'   predictors. See the \strong{Details} and \strong{Examples} sections.
+#'   package.
 #' @param xbar A numeric vector of column means in the implicit design matrix 
 #'   excluding the intercept for the observations included in the model.
 #' @param ybar A numeric scalar indicating the mean of the outcome for the

dev-notes/rstanarm_dev_notes.md

@@ -43,7 +43,7 @@ Regarding the above diagram, the `model_*.R` and `model.fit` files are in the `R
 
 In the .fit function you should,
 
-* Center the covariates.
+* Center the covariates (if there is an intercept).
 * Perform QR decomposition.
 * Determine whether the intercept is declared (or whether multiple linear predictors have been declared).
 * Extract and process information regarding the priors declared.
@@ -66,12 +66,12 @@ Conditionals should be set up so that the .fit workhorse function can deal with
 
 In the Stan file you should try to minimize the number of loops, storing n-dimensional objects, redoing calculations that only need to be done once, calculating inverses (e.g. use the precision multinormal instead of multinormal).
 
-Because we center the covariates you have to separate the intercept out of the linear predictor matrix (i.e. `X` should not contain a vector of ones). If `gamma` is the intercept parameter fit using centered predictors and `alpha` is the intercept parameter you want to report then do the following transformation in generated quantities:
+Because we center the covariates when there is an intercept, you have to separate the intercept out of the linear predictor matrix (i.e. `X` should not contain a vector of ones). If `gamma` is the intercept parameter fit using centered predictors and `alpha` is the intercept parameter you want to report then do the following transformation in generated quantities:
 ```
 ...
 generated quantities {
   real alpha[has_intercept];
-  {
+  if (has_intercept) {
     alpha[1] = gamma[1] - dot_product(beta, xbar)
   }
 }

man-roxygen/args-priors.R

@@ -37,4 +37,8 @@
 #'   \strong{Note:} If using a dense representation of the design matrix 
 #'   ---i.e., if the \code{sparse} argument is left at its default value of
 #'   \code{FALSE}--- then the prior distribution for the intercept is set so it
-#'   applies to the value when all predictors are centered.
+#'   applies to the value when all predictors are centered. If you prefer to
+#'   specify a prior on the intercept without the predictors being auto-centered,
+#'   then you have to omit the intercept from the \code{\link[stats]{formula}}
+#'   and include a column of ones as a predictor, in which case some element
+#'   of \code{prior} specifies the prior on it, rather than \code{prior_intercept}.

tests/testthat/helpers/expect_equivalent_loo.R

@@ -1,4 +1,5 @@
 expect_equivalent_loo <- function(fit) {
+  options(loo.cores = ifelse(.Platform$OS.type == "windows", 1, 2))  
   l <- suppressWarnings(loo(fit))
   w <- suppressWarnings(waic(fit))
   expect_s3_class(l, "loo")

tests/testthat/test_loo.R

@@ -16,7 +16,7 @@
 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
 
 library(rstanarm)
-options(loo.cores = 2)
+options(loo.cores = ifelse(.Platform$OS.type == "windows", 1, 2))
 SEED <- 1234
 set.seed(SEED)
 CHAINS <- 2

vignettes/lm.Rmd

@@ -340,10 +340,8 @@ output of the `biglm` function in the __biglm__ package, which utilizes an
 incremental QR decomposition that does not require the entire dataset to be
 loaded into memory simultaneously. However, the `biglm` function needs to be
 called in a particular way in order to work with `stan_biglm`. In particular,
-the columns of the implicit design matrix need to be centered, as does the
-outcome, in which case the formula must _not_ include an intercept. The means
-of the columns of the design matrix, the sample mean of the outcome, and the
-sample standard deviation of the outocme all need to be passed to the `stan_biglm` 
+The meansof the columns of the design matrix, the sample mean of the outcome, and 
+the sample standard deviation of the outocme all need to be passed to the `stan_biglm` 
 function, as well as a flag indicating whether the model really does include an
 intercept. Also, the number of columns of the design matrix currently cannot 
 exceed the number of rows. Although `stan_biglm` should run fairly quickly and

vignettes/priors.Rmd

@@ -90,7 +90,7 @@ leave those arguments at their defaults (i.e., we don't specify them):
 
 ```{r, default-prior-1, results="hide"}
 library("rstanarm")
-default_prior_test <- stan_glm(mpg ~ wt + am, data = mtcars)
+default_prior_test <- stan_glm(mpg ~ wt + am, data = mtcars, chains = 1)
 ```
 
 The `prior_summary` function provides a concise summary of the priors used: 
@@ -129,7 +129,12 @@ when $x=0$ is not meaningful and it is easier to think about the value when $x =
 predictors typically makes it easier to specify a reasonable prior for the
 intercept. (Note: the user does _not_ need to manually center the predictors.)
 
-The next two subsections describe how this rescaling works and how to easily
+To disable the centering of the predictors, you need to omit the intercept from
+the model `formula` and include a column of ones as a predictor (which cannot
+be named `"(Intercept)"` in the `data.frame`). Then you can specify a prior
+"coefficient" for the column of ones.
+
+The next two subsections describe how the rescaling works and how to easily
 disable it if desired.