2 IAT: Sexuality

Author

Affiliation

Seungju Kim

University of Illinois at Urbana-Champaign

Source Description

Full name: Implicit Association Test - Sexuality
Website: Harvard Project Implicit
Data type: Survey
Unit of analysis: State
Geographic Coverage: 50 U.S. states + D.C.
Temporal Coverage: 2016–2025

About the Original Data

Project Implicit (housed within Harvard University) is a web-based platform that collects implicit associations and explicit attitudes toward sexuality. Data collection for the sexuality IAT has been ongoing since 2004 using a non-probability sample of volunteer participants who visit the website through search engines, social media, or school/work referrals.

Warning

Researchers should be aware of one significant limitation of the implicit associations measure for sexuality: it only captures implicit associations between good/bad and gay men (via the D-score). This means the implicit measure may not hold meaningful implications for lesbian, bi/pansexual, or otherwise sexually diverse individuals.

Original Data Collection

Feature	Details
Collection method	Continuous online survey (self-selected)
Sampling frame	U.S. adult internet users
Sample type	Convenience (non-probability)
Update frequency	Annual public data releases
Geographic coverage	State-level (U.S. only)
Temporal coverage	2004–present

Note

Project Implicit releases public-use datasets annually. File formats vary by year — earlier years are released as .sav (SPSS), later years as .csv. Variable naming conventions also shift across years (e.g., tgaywomen → tgayleswomen starting in 2017).

Required Citation

If you use this indicator in published research, you must cite the original data source in addition to the stigmaR package:

Xu, K., Nosek, B., & Greenwald, A. (2014). Psychology data from the Race Implicit Association Test on the Project Implicit Demo website. Journal of Open Psychology Data, 2(1). https://doi.org/10.5334/jopd.ac

Access and Download

Feature	Details
Access type	Public download
URL	OSF — Project Implicit Public Data
Authentication required	No
Cost	Free
Last accessed	January 2026

Download Procedure

Annual sexuality IAT datasets are available on the Project Implicit OSF page. Files are organized by year and released in .csv or .sav format.

Navigate to the OSF repository linked above.
Download one file per year for the Sexuality IAT (2016–2025).
Place files in data/raw_data/iat/sexuality/, organized by year.

Warning

2015 data uses a different variable set — it is missing several policy items (adoptchild, serverights, transgender, marriagerights_3num, relationslegal_3num) that are required for the composite indices. 2015 is excluded from this pipeline. Pre-2015 data also differ substantially in structure and are not used.

Cleaning Procedure

Raw Data Structure

File format: .csv (2016+) and .sav (select years)
Rows: One row per respondent
Key variables used: state, year, d_biep.straight_good_all, att_7, tgaymen, tgayleswomen / tgaywomen, marriagerights_3num, relationslegal_3num, adoptchild, serverights, transgender

Cleaning Steps

1. File Loading and Year Exclusion

All .csv and .sav files in the raw data folder are read in and merged into a single named list. The 2015 file is excluded due to incompatible variable structure.

csv_files <- csv_files[!grepl("2015", csv_files)]

Column names are normalized to lowercase to handle inconsistent casing across years and file types.

names(df) <- tolower(names(df))

2. Variable Construction and Scoring

Nine item-level variables are constructed, each scored so that higher values = more stigma (anti-gay/lesbian attitudes).

A helper function grab() is used throughout to safely retrieve columns that may not exist in all years — returning NA if the variable is absent rather than throwing an error.

grab <- function(df, col) {
  if (col %in% names(df)) df[[col]] else rep(NA_real_, nrow(df))
}

Installing package into '/home/runner/work/_temp/Library'
(as 'lib' is unspecified)

also installing the dependency 'remotes'

Installing package into '/home/runner/work/_temp/Library'
(as 'lib' is unspecified)

also installing the dependencies 'Rcpp', 'lazyeval', 'later', 'otel', 'crosstalk', 'htmlwidgets', 'magrittr', 'promises'


DT installed

Important

iat_sex_explicit_bel (belief that sexuality is environmentally determined, from sexualityorigin) was planned but is not included — this variable does not appear in any Project Implicit sexuality dataset from 2016 onward.

3. State Filtering

After aggregation, state strings are normalized (uppercase, whitespace stripped, SPSS labels removed via haven::zap_labels()). Only the 50 U.S. states + D.C. are retained.

mutate(state = toupper(trimws(as.character(haven::zap_labels(state))))) |>
  filter(state %in% us_states_dc)

Aggregation to State Level

Unit of Analysis in Raw Data

Individual respondent (one row per person per year).

Aggregation Method

Feature	Decision	Rationale
Statistic	Mean	Standard for IAT D-score and Likert items aggregated to state level
Weighting	Unweighted	Project Implicit does not provide sampling weights
Missing states	Excluded	States with no respondents in a given year produce no row

For each item, both the mean and a respondent count (iat_sex_n_{item}) are computed per state × year. The count columns are retained in the items dataset for downstream quality checks.

group_by(year, state) |>
  summarise(
    across(c(iat_sex_imp_d, ...), ~ sum(!is.na(.x)), .names = "iat_sex_n_{sub('iat_sex_', '', .col)}"),
    across(c(iat_sex_imp_d, ...), ~ mean(.x, na.rm = TRUE), .names = "{.col}"),
    .groups = "drop"
  )

Output Dataset

Variables in Final Output

Item-Level

iat_sexuality_items.Rds — item-level dataset (state × year)

Variable	Type	Description	Scoring
`state`	chr	Two-letter state abbreviation	—
`year`	int	Calendar year	2016–2025
`iat_sex_imp_d`	dbl	Implicit D-score	Higher = more pro-straight bias
`iat_sex_exp_att`	dbl	Explicit attitude (7-pt)	Higher = more stigma
`iat_sex_exp_therm_gm`	dbl	Thermometer: gay men (reversed)	Higher = more stigma
`iat_sex_exp_therm_gw`	dbl	Thermometer: lesbian women (reversed)	Higher = more stigma
`iat_sex_exp_pol_marr`	dbl	Policy: marriage equality	0–1; higher = more stigma
`iat_sex_exp_pol_legal`	dbl	Policy: legal recognition	0–1; higher = more stigma
`iat_sex_exp_pol_adopt`	dbl	Policy: adoption rights	0–1; higher = more stigma
`iat_sex_exp_pol_serv`	dbl	Policy: right to refuse service	0–1; higher = more stigma
`iat_sex_exp_pol_trans`	dbl	Policy: trans bathroom access	0–1; higher = more stigma
`iat_sex_n_{item}`	int	Respondent count per item	—

Composite-level

iat_sexuality_indices.Rds — composite-level dataset (state × year)

Variable	Type	Description	Components
`state`	chr	Two-letter state abbreviation	—
`year`	int	Calendar year	2016–2025
`iat_sex_implicit`	dbl	Implicit stigma index	`iat_sex_imp_d`
`iat_sex_explicit_therm`	dbl	Thermometer composite	Mean of `therm_gm`, `therm_gw`
`iat_sex_explicit_pol`	dbl	Policy composite	Mean of 5 policy items
`iat_sex_explicit`	dbl	Omnibus explicit index	Mean of all 7 explicit items

Composites use rowMeans(..., na.rm = TRUE) so that state-years with partial item coverage still receive a score.

Output Files

saveRDS(iat_sexuality_items,   here("data/clean_data/iat_sexuality_items.Rds"))
saveRDS(iat_sexuality_indices, here("data/clean_data/iat_sexuality_indices.Rds"))

File locations:

data/clean_data/iat_sexuality_items.Rds
data/clean_data/iat_sexuality_indices.Rds

Known Limitations

Non-probability sample: Project Implicit data come from self-selected volunteers. State-level estimates reflect the online population visiting the site, not representative state samples.
Implicit measure scope: The D-score measures implicit associations for gay men only. Implicit attitudes toward lesbian women, bisexual, or other sexually diverse groups are not captured.
iat_sex_explicit_bel not available: The planned belief-about-origin item (sexualityorigin) does not appear in any 2016+ dataset and is excluded.
2015 excluded: Variable structure is incompatible with 2016+; a separate pipeline would be needed to incorporate it.
Thermometer variable rename: tgaywomen was renamed tgayleswomen starting in 2017. The pipeline handles this via coalesce() but researchers should verify continuity if comparing pre- vs. post-2017 estimates.