Skip to contents

Inference for Nonprobability Samples Using Multiple Reference Surveys

nonprobsampling implements pseudo-weighting methods for finite population inference from nonprobability samples, such as convenience samples, volunteer cohorts, and opt-in panels. Because the participation mechanism in a nonprobability sample is unknown, unweighted estimates of population means and prevalence may be biased. The package addresses this issue by leveraging auxiliary information from one or multiple probability reference surveys to estimate participation probabilities and using their inverses as pseudo-weights to obtain estimates of finite population means and prevalence adjusted for participation bias.

The extent of bias reduction depends on the assumptions underlying the chosen method and the auxiliary information available.

The implemented methods are based on the generalized estimating equations framework of Landsman et al. (2026). This includes a multi-reference extension that enables the integration of auxiliary information across multiple reference surveys.

Installation

Install the released version from CRAN:

install.packages("nonprobsampling")

Install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("Jiakun0611/nonprobsampling")

Methods

With one reference survey:

  • Calibration — raking ratio calibration (Landsman et al., 2026)
  • Adjusted logistic propensity (ALP) weighting method — Wang, Valliant, and Li (2021)
  • CLW — Chen, Li, and Wu (2020)

With multiple reference surveys:

  • Multi-reference calibration — enables the integration of auxiliary information across multiple surveys when no single reference survey contains all variables relevant to participation (Landsman et al., 2026), with an optional cumulative precalibration step to preserve information on the relationships between overlapping and survey-specific auxiliary variables across reference surveys.

Variance estimation is based on Taylor linearization, with complex sampling designs in the reference surveys handled through integration with the survey package. When a reference survey is supplied as a replicate-weight design, its contribution to the variance is estimated using the supplied replicate weights.

Usage

Estimation proceeds in two steps: est_pw() estimates pseudo-weights, then pwmean() estimates a pseudo-weighted mean or prevalence for an outcome.

For user-supplied data, harmonize shared variables before creating survey design objects. Categorical variables must have identical factor levels in the same order and the same contrast coding across samples; treatment contrasts must use the same reference category. The package checks factor levels but does not check contrast coding. See the variable harmonization guidance for details.

library(nonprobsampling)

data(sc)    # nonprobability sample (outcome: psa_level)
data(sp1)   # probability reference survey

# Reference survey design
ref1_design <- survey::svydesign(
  ids     = ~psu_sp1,
  strata  = ~strata_sp1,
  weights = ~wts_sp1,
  data    = sp1,
  nest    = TRUE
)

# Step 1: estimate pseudo-weights (one-reference calibration)
fit <- est_pw(
  data      = list(sc, ref1_design),
  p_formula = ~ agecat + race + education + comorbidity + BMI + diabetes,
  method    = "calibration",
  control   = pw_solver_control(ftol = 1e-6)
)
print(fit)
summary(fit)

# Step 2: pseudo-weighted mean of the outcome, by BMI categories
out <- pwmean(fit, y = "psa_level", zcol = "BMI")
print(out)
summary(out)

With multiple reference surveys, users provide one survey design object for each reference survey and a corresponding list of participation model formulas, with one formula specified for each survey.

data(sp2)   # second probability reference survey

ref2_design <- survey::svydesign(
  ids     = ~psu_sp2,
  strata  = ~strata_sp2,
  weights = ~wts_sp2,
  data    = sp2,
  nest    = TRUE
)

fit2 <- est_pw(
  data = list(sc, ref1_design, ref2_design),
  p_formula = list(
    ~ agecat + race + education + psa_level + pros_enlarged + comorbidity,
    ~ agecat + race + BMI + diabetes + comorbidity
  ),
  sp_order = "size",
  precali = TRUE,
  control = pw_solver_control(ftol = 1e-6)
)
print(fit2)
summary(fit2)

See vignette("nonprobsampling") for more details.

Datasets

The package includes example datasets used throughout the documentation:

Dataset Description
sc Nonprobability sample (synthetic, NHANES-based)
sp1 First probability reference survey (NHANES 1999–2010)
sp2 Second probability reference survey (NHIS 1997–2008)
sp1_bootstrap sp1 with bootstrap replicate weights

Getting help

Click any topic to open its help page (or run the command in R):

## Package overview
?nonprobsampling

## Main functions
?est_pw
?pwmean
?pw_solver_control

## Datasets
?sc
?sp1
?sp2
?sp1_bootstrap

## Vignette
vignette(“nonprobsampling”)

Citation

To cite the package in publications, run:

citation("nonprobsampling")

References

  • Chen, Y., Li, P., and Wu, C. (2020). Doubly robust inference with nonprobability survey samples. Journal of the American Statistical Association, 115(532), 2011–2021. doi:10.1080/01621459.2019.1677241
  • Landsman, V., Wang, L., Carrillo-Garcia, I., Mitani, A. A., Smith, P. M., Graubard, B. I., Bui, T., and Carnide, N. (2026). Correction for participation bias in nonprobability samples using multiple reference surveys. Statistics in Medicine, 45(3–5). doi:10.1002/sim.70403
  • Wang, L., Valliant, R., and Li, Y. (2021). Adjusted logistic propensity weighting methods for population inference using nonprobability volunteer-based epidemiologic cohorts. Statistics in Medicine, 40(24), 5237–5250. doi:10.1002/sim.9122

License

GPL-3 © Jiakun Lin, Victoria Landsman, Aya A. Mitani