simmr_elicit: Function to allow informative prior distribution to be...

Description Usage Arguments Details Value Author(s) Examples

View source: R/simmr_elicit.R

Description

The main simmr_mcmc function allows for a prior distribution to be set for the dietary proportions. The prior distribution is specified by transforming the dietary proportions using the centralised log ratio (CLR). The simmr_elicit and simmr_elicit functions allows the user to specify prior means and standard deviations for each of the dietary proportions, and then finds CLR-transformed values suitable for input into simmr_mcmc.

Usage

1
2
3
4
5
6
simmr_elicit(
  n_sources,
  proportion_means = rep(1/n_sources, n_sources),
  proportion_sds = rep(0.1, n_sources),
  n_sims = 1000
)

Arguments

n_sources

The number of sources required

proportion_means

The desired prior proportion means. These should sum to 1. Should be a vector of length n_sources

proportion_sds

The desired prior proportions standard deviations. These have no restricted sum but should be reasonable estimates for a proportion.

n_sims

The number of simulations for which to run the optimisation routine.

Details

The function takes the desired proportion means and standard deviations, and fits an optimised least squares to the means and standard deviations in turn to produced CLR-transformed estimates for use in simmr_mcmc. Using prior information in SIMMs is highly desirable given the restricted nature of the inference. The prior information might come from previous studies, other experiments, or other observations of e.g. animal behaviour.

Due to the nature of the restricted space over which the dietary proportions can span, and the fact that this function uses numerical optimisation, the procedure will not match the target dietary proportion means and standard deviations exactly. If this problem is severe, try increasing the n_sims value.

Value

A list object with two components

mean

The best estimates of the mean to use in control.prior in simmr_mcmc

sd

The best estimates of the standard deviations to use in control.prior in simmr_mcmc

Author(s)

Andrew Parnell <andrew.parnell@mu.ie>

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
## Not run: 
# Data set: 10 observations, 2 tracers, 4 sources
data(geese_data_day1)
simmr_1 <- with(
  geese_data_day1,
  simmr_load(
    mixtures = mixtures,
    source_names = source_names,
    source_means = source_means,
    source_sds = source_sds,
    correction_means = correction_means,
    correction_sds = correction_sds,
    concentration_means = concentration_means
  )
)

# MCMC run
simmr_1_out <- simmr_mcmc(simmr_1)

# Look at the prior influence
prior_viz(simmr_1_out)

# Summary
summary(simmr_1_out, "quantiles")
# A bit vague:
#           2.5%   25%   50%   75% 97.5%
# Source A 0.029 0.115 0.203 0.312 0.498
# Source B 0.146 0.232 0.284 0.338 0.453
# Source C 0.216 0.255 0.275 0.296 0.342
# Source D 0.032 0.123 0.205 0.299 0.465

# Now suppose I had prior information that:
# proportion means = 0.5,0.2,0.2,0.1
# proportion sds = 0.08,0.02,0.01,0.02
prior <- simmr_elicit(4, c(0.5, 0.2, 0.2, 0.1), c(0.08, 0.02, 0.01, 0.02))

simmr_1a_out <- simmr_mcmc(simmr_1, prior_control = list(means = prior$mean, sd = prior$sd))

#' # Look at the prior influence now
prior_viz(simmr_1a_out)

summary(simmr_1a_out, "quantiles")
# Much more precise:
#           2.5%   25%   50%   75% 97.5%
# Source A 0.441 0.494 0.523 0.553 0.610
# Source B 0.144 0.173 0.188 0.204 0.236
# Source C 0.160 0.183 0.196 0.207 0.228
# Source D 0.060 0.079 0.091 0.105 0.135

## End(Not run)

simmr documentation built on Feb. 27, 2021, 5:05 p.m.