Frequently Asked Questions
1. Can I plot more than 1000 ages?
The default number of rows in Radial/DensityPlotter is
capped at 1000. To process larger datasets, please prepare a .csv
input file as explained in this section for
Fission Track data, and in
this section for Other data.
2. How many peaks to fit?
The algorithms of Galbraith and Green (1993) assume that the true
age distribution underlying the data consists of a number of
discrete peaks. This assumption is necessary for reasons of parsimony.
In reality, however, discrete events rarely occur in Nature. It is
important to keep this in mind when using automated algorithms such
as the Bayes Information Criterion.
Given enough data, criteria like this inevitably provide statistical
justification to fit more peaks. It is generally more informative
to use a fixed number of peaks or apply the minimum age model.
3. Can I change the bandwidth of my KDE?
DensityPlotter uses the algorithm of Botev et al. (2010)
to set a default bandwidth. This is just one of literally dozens
of similar bandwidth selection algorithms, which seems to work
quite well for geochronological data. Sometimes, however, the
Botev-algorithm fails to converge, leading
DensityPlotter to fall back on a simpler algorithm.
This tends to happen more often for small datasets, and produces
very smooth and rather uninformative KDE curves. Another situation
in which one might want to change the default bandwith is when
comparing multiple age distributions from one and the same dataset.
Visual comparison of such multi-panel diagrams is much easier
when a common bandwidth is used. DensityPlotter
allows the user to manually adjust the bandwidth of KDEs as well
as the binwidth of histograms to account for these situations
in the Settings form.
In such situations, the user is advised to report the bandwidth
in the figure caption.
4. Why don't the peaks in my mixture model fit the modes?
First, the number modes in a KDE strongly depends on the kernel bandwidth.
Changing this to a higher or lower value may lead to spurious modes that
do not necessarily reflect true underlying populations.
Second, the mixture modelling uses the algorithms of
Galbraith and Green (1993), which have been adopted by Sambridge and
Compston (1994) or U-Pb dating as well. When applied to Other
data, the model assumes lognormal residuals. This assumption may or may
not be valid for the data.