Rich: Your use of the term MANOVA suggests a multivariate response (Y).
If what you really have is multiple factors (predictors), then this is a
different modeling construct (multiple regression) and it would seem
nonpartest() is not appropriate. I've been analyzing water quality
constituents (one at a time as a univariate response) with multiple
predictors (I'm using years, seasons within years, stream flow, location
within watershed but anything you might include in a regression model could
be included) in linear quantile regression (quantreg package). This is a
semiparametric approach in the sense that I don't have to make any
assumption about a particular distributional form of the error distribution
since estimating the quantiles is estimating the inverse of an empirical
cumulative distribution, but all the predictors have parameters associated
with them. You can include contrasts for categorical predictors,
interactions, etc. You also could relax the linear additive model by using
smoothing functions (e.g., b-splines) on the predictors.
Brian
Brian S. Cade, PhD
U. S. Geological Survey
Fort Collins Science Center
2150 Centre Ave., Bldg. C
Fort Collins, CO 80526-8818
email: cadeb@usgs.gov <brian_cade@usgs.gov>
tel: 970 226-9326
On Mon, May 12, 2014 at 5:03 PM, Rich Shepard
<rshepard@appl-ecosys.com>wrote:
> I read a short thread on this topic from last June on stackoverflow.com.
> Both Bryan Hanson and Ben Bolker suggested looking for such functions using
> the sos package. I did this; the nonpartest function in npmv does not look
> to me like it does what I need. Since that thread did not reach a
> definitive
> conclusion (no response by the original poster) I would like to continue my
> search here.
>
> Context: water quality constituent concentrations measured multiple times
> from each of several monitoring wells with the latter in two groups: up-
> and
> down-gradient from a site of potential ground water contamination.
>
> To compare variation within individual monitoring wells with variation
> between the wells I'd use the Kruskal-Wallis test. That would also be
> appropriate to compare variation within each group (up-gradient and
> down-gradient) with variation between the two groups. My understanding of
> multiple analysis of variance is this would allow multiple sources of
> variability (intra-well and between wells) as explanatory variables for the
> response variable of a specified chemical constituent.
>
> The help page for nonpartest() tells me that it analyzes one-way
> multivariate data. Perhaps my understanding of this is poor, but the help
> page tells me that nonpartest() formula has a single explanatory variable
> and multiple response variables. In the context above, I have a single
> response variable (a chemical constituent concentration) and multiple
> explanatory variables (at least date, well, and location).
>
> Please confirm that my understanding of nonpartest() is correct and
> suggest an appropriate R protocol to analyze concentration variability
> within monitoring wells and between two well locations (up- and
> down-gradient).
>
> TIA,
>
> Rich
>
> ______________________________________________
> R-help@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/
> posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
[[alternative HTML version deleted]]