Hakuna my data

R codes and data accompanying the paper "Understanding the reliability of citizen science observational data using item response models" published in Methods in Ecology and Evolution.

Authors:

Edgar Santos-Fernandez, PhD. School of Mathematical Sciences. Y Block, Floor 8, Gardens Point Campus Queensland University of Technology. GPO Box 2434. Brisbane, QLD 4001. Australia

Kerrie Mengersen, Dist. Professor of Statistics. Deputy Director, ARC Centre for Mathematical and Statistical Frontiers; Director, QUT Centre for Data Science. School of Mathematical Sciences. Y Block, Floor 8, Gardens Point Campus. Queensland University of Technology. GPO Box 2434. Brisbane, QLD 4001. Australia

Preprint versions of the paper can be found at:

The models introduced in the article are part of the R package staircase (https://github.com/EdgarSantos-Fernandez/staircase).

Content:

There are two RMarkdown files:

simple_example: shows the fit of a spatial item response model using a small dataset (for illustration purposes only and these results do not appear in the paper). There is a HTML and a PDF version.
Item_response_modeling_for_big_data: contains the codes used to fit the models using big datasets on a High-Performance Computing system (Paper section 4.3). There is a HTML output from this file.

Some outputs from the paper:

Fig: Estimates of the users' abilities vs their proportion of correct classification.

Fig: Species difficulties vs the proportion of correct classification.

Item_response_modeling_for_big_data:

1- Codes_used_in_Item_response_modeling_for_big_data.Rmd illustrate the implementation of the case study from Section 4.3 (Item response model for big data). The first part contains the codes for fitting the model using Stan on a High-Performance Computing system (HPC). We then show how to combine the subposterior estimates via consensus Monte Carlo.

2- serengeti.RDS is the dataset used in the case study. It includes the following variables:

LocationX and LocationY: X and Y coordinates of the site.
DateTime: classification time stamp.
UserID: unique user identifier.
SiteID: site identification code.
Species: classification value.
True_Species: true species obtained from the gold standard dataset or via consensus voting.
Correct: outcome of the classification (1 = correct, 0 = otherwise).
set: Data was split into 10 shards using stratified sampling.

See Swanson et al. (2015) for more details.

How to cite:

Santos-Fernandez, Edgar, and Mengersen, Kerrie. "Understanding the reliability of citizen science observational data using item response models". Methods in Ecology and Evolution.
Santos-Fernandez, Edgar, and Mengersen, Kerrie. "Bayesian item response models for citizen science ecological data." arXiv preprint arXiv:2003.06966 (2020).

References: