Skip to contents

08: Initial Values

RSTr-initialvalues.Rmd

Overview

inits is a list specifying the starting values for parameters in the model. By default, all of these values are specified according to the literature, but RSTr allows the capability of specifying your own initial values. If you wish to provide initial values, note that you don’t have to specify inits for all parameters if you only want to specify some of them - any undefined inits will be defined by the default values. For example, you can specify only the initial values for lambda and all other values will be generated on their own. However, if one value is specified for a certain parameter in inits, all values must be specified for that parameter in inits: you cannot, for example, define initial values for just one year of lambda. Finally, any values included in your inits list that aren’t aligned with the above names will be ignored.

Initial value specifications

The models in RSTr share many inits, but a couple of models have inits that are unique to them. All potential initial values are presented here.

Initial values for the MSTCAR model

Here are the possible initial value parameters for the MSTCAR model:

  • lambda: The estimated spatially smoothed rate for each region-group-time. lambda is an array of real numbers with dimensions n_region x n_group x n_time. Has support (0, 1) for method = "binomial" and support (0, Inf) for `method = “poisson”;

  • beta: The mean rate for each island-group-year on a logit- or log-transformed scale. Islands are sets of regions that exclusively share adjacency information. For example, in miadj, there are two islands that represent the counties of the Upper Peninsula and the Lower Peninsula. These islands don’t touch each other, and thus don’t share adjacency information. Each island is assigned its own beta. beta is an array of real numbers with dimensions n_island x n_group x n_time;

  • Z: The spatiotemporal random effects. These are the parameters that induce smoothing on the counties, with the intensity of the smoothing dictated by the spatial covariance matrices G. Z is an array of real numbers with dimensions n_region x n_group x n_time;

  • G: The spatial covariance matrices. This parameter determines the intensity of the spatial smoothing performed by Z and represents the strength of the relationship between each group in a given time period. G is an array of temporally-evolving positive-definite symmetric matrices with dimensions n_group x n_group x n_time;

  • rho: The temporal correlation. This parameter decides the strength of the relationship between values in time period t to values in time period t-1. It is a matrix of size n_group x 1 of real numbers with support [0,1];

  • tau2: The non-spatial variance. This parameter picks up any variance in values of lambda for each group. It is a matrix of size n_group x 1 of positive real numbers; and

  • Ag: The general spatial covariance matrix. This parameter describes the overall relationship between groups across the entire model and is used in the prior distribution for the matrices in G. Ag is a positive-definite symmetric matrix with dimensions n_group x n_group.

Initial values for the MCAR model

The MCAR model utilizes a majority of the initial values of the MSTCAR model. However, MCAR does not include inits for rho or Ag. Note that specification for the MCAR model is slightly different than that of the MSTCAR model. If an MCAR model is run with data containing several time periods, tau2 will require values for every time period along with every group.

Initial values for the CAR/RCAR model

The CAR models have the smallest set of initial values, using only lambda, beta, Z, and tau2 from the MCAR model. Similar to the MCAR, if a CAR model is run with multiple groups and time periods, tau2 requires values for every group and time period present. The only new initial value for the CAR models is sig2, which takes the place of G in the MCAR and MSTCAR models:

  • sig2 represents the spatial variance of a CAR/RCAR model. This parameter picks up any variance in values of Z for each group. It is a matrix of size n_group x n_time of positive real numbers.