v0.2.0

Features and functions

• New option to skip calculation of Bayesian estimates in PoolPrev(), substantially speeding up this function.
• New default function values for PoolPrev() and HierPoolPrev() to improve point estimates.
  • New robust parameter has default value robust = TRUE, which means the point estimate of prevalence is the posterior median.
  • Default value for all.negative.pools parameter isall.negative.pools = 'zero', meaning when all pools are negative, the point estimate and the lower bound for the interval will be 0.
• Removed one source of bias from prevalence estimates returned for any hierarchical models, impacting HierPoolPrev() and getPrevalence() output.
• Random effects are marginalised out when calculating population-level prevalence.
  • We no longer support specifying nested surveys using ~(1|Layer1/Layer2)
  • We recommend using the format ~(1|Layer1) + (1|Layer2), which should be equivalent as long as each level in Layer2 is unique
• HierPoolPrev() now returns estimate of intracluster correlation coefficients (ICC) at one or more levels of clustering.
• See the NEWS.md file for full details on what's changed

Development

• Updating github actions by @njtierney in #12
• Update deprecated syntax for future rstan compatibility by @andrjohns in #14

New Contributors

• @njtierney made their first contribution in #12
• @andrjohns made their first contribution in #14
• @caitlinch made their first contribution in #22

Full Changelog

v0.1.3...v0.2.0

PoolTestR 0.1.3

This is patch to fix a bug affecting PoolPrev. The bug affected the maximum likelihood estimates (MLE) and likelihood ratio confidence intervals (LR-CIs) of prevalence when the default Jeffrey's prior was being used. The bug would usually make the MLE and LR-CIs much closer to the Bayesian estimates than they should have been. As both sets of estimates are valid, the results will still have been approximately correct.

This patch also includes an option, replicate.poolscreen (default to FALSE), for PoolPrev. This options changes the way the likelihood ratio confidence intervals are calculated. With replicate.poolscreen = TRUE PoolPrev will more closely reproduce the results produced by Poolscreen. We believe that our implementation of these intervals is more correct so would recommend that users continue to use the default (replicate.poolscreen = FALSE), but this option may be helpful for those who are trying to compare results across the two programs.

v0.1.1

This is the first release that compiles on all platforms supported by CRAN (previously, the package failed on Solaris)

v0.1.0

First version available on CRAN