Internal package to standardize preliminary data analysis and integrate analytical feedback
Encodes common statistical decision trees, for a list of response variables for a project.
Cite via Zenodo DOI --
Requires input data tbl to be "nested" into 2 columns --
- "variable", a vector of response variable names, and
- "varData", a list of
This can be done with the 2 code lines, acting on typically cleaned "wide" data (i.e. response variables as columns), with dataset's independent variable keys also as columns (e.g. "plot", treatment", etc.) --
#e.g. cleanData <- dplyr::tibble( "plot" = c(1, 2, 3), "treat" = c(1, 2, 3), "var1" = c(1, 2, 3), "var2" = c(1, 2, 3) #, [...] )
varTbl <- cleanData %>% tidyr::pivot_longer(names_to = "variable", values_to = "value") %>% purr::nest(cols = !c(-variable)).
then processed using --
statFormula <- value ~ [independent variable of interest] resultsTbl <- varTbl %>% oir::getStatsTbl(formula = statFormula)
Notes --
- best used with simple model formula, i.e. 1 independent variable in "statFormula" (at a time)
- the main oir::getStatsTbl() function has versions "1", "2", and "12" to run --
- linear regressions at plot-/group-level median centers ("1")
- non-linear version of regression (not nlme::) modifying independent variable with stats::poly(x, degree = 2, at finest row level ("2")
- the same non-linear version of regression, but using plot-/group-level centers ("12")
- more complex formulas (e.g. lmer) can be processed, but with limited insight into final appropriate p-values
- non-parametric option included as back-up test for usable p-value, but may offer slightly different inference/interpretation