test.Maf: Test number of significant maf time series in data

In matzhaugen/maf: Maximum Autocorrelation Factors

Description

Perform a statistical test for how many signals are hiding in the data. The output gives p-values for each time series

Usage

test.Maf(maf.obj, alpha = 0.05, block.size = 5, smooth.span = 30,
  B = 100)

Arguments

alpha	The significance level
block.size	The block size of the resampled residuals, i.e. the number of contiguous time steps to sample at the time from the set of residuals.
smooth.span	Fraction between 0-1 that specifies the proportion of timesteps to include in the smoothing window, which is weighted by a tricubic. See ?loess for details. Alternatively, one can specify an integer greater than 1 which refers to the number of time points to include in the tricubic filter.
B	Number of replications in the confidence interval calculations
maf.object	The output of the maf function.

Details

The confidence intervals are obtained using a resampling scheme which extracts the residuals of the original time series after smoothing with the same filter as the one parametrized by smooth.span. The residuals are bootstrapped or block bootstrapped (if there is temporal structure in the residuals) and added back to the smooth, creating a new data set. From the new data set a new set of MAFs are calculated.

The smoothing parameter smooth.span is also used when calculating the empirical Signal-to-Noise ratio. Where we subtract the smooth from the time series and calculate the variance of the smooth estimate over the residual variance. This gives an empirical signal-to-noise estimate.

Value

A list with the following items

statStar

Empirical SNR for each resampled MAF time series, i.e. a p x B matrix where p is the total number of predictors and B is the number of bootstraps/replications

statObs

The empirical SNR of the original MAF time series.

pval

The p-values of each MAF, where the p-value refers to the number of resampled MAFs that have a higher empirical SNR than the original MAFs. If there are less than α, e.g. 0.05, resampled MAFs that have a higher empirical SNR than the original MAF, the MAF in question is not significant.

mafStar

The B resampled maf timeseries, in a 3d array.

mafSmooth

The smooth original maf estimates.

Examples

# Extract mafs from dataset
maf.object = maf(treeringTimeseries)
# Test for how many mafs are present
test.Maf(maf.object) 

matzhaugen/maf documentation built on May 21, 2019, 1:36 p.m.