nevada - Network-Valued Data Analysis
A flexible statistical framework for network-valued data
analysis. It leverages the complexity of the space of
distributions on graphs by using the permutation framework for
inference as implemented in the 'flipr' package. Currently,
only the two-sample testing problem is covered and
generalization to k samples and regression will be added in the
future as well. It is a 4-step procedure where the user chooses
a suitable representation of the networks, a suitable metric to
embed the representation into a metric space, one or more test
statistics to target specific aspects of the distributions to
be compared and a formula to compute the permutation p-value.
Two types of inference are provided: a global test answering
whether there is a difference between the distributions that
generated the two samples and a local test for localizing
differences on the network structure. The latter is assumed to
be shared by all networks of both samples. References: Lovato,
I., Pini, A., Stamm, A., Vantini, S. (2020) "Model-free
two-sample test for network-valued data"
<doi:10.1016/j.csda.2019.106896>; Lovato, I., Pini, A., Stamm,
A., Taquet, M., Vantini, S. (2021) "Multiscale null hypothesis
testing for network-valued data: Analysis of brain networks of
patients with autism" <doi:10.1111/rssc.12463>.
