Applies nonresponse weighting to a creel_design object by scaling
survey weights within each stratum by the inverse of the observed response
rate. The adjustment uses postStratify (default) or
calibrate to update the internal
svydesign object, so all downstream estimators
(estimate_effort, estimate_catch_rate, etc.)
automatically use the corrected weights.
Usage
adjust_nonresponse(
design,
response_rates,
method = c("postStratify", "calibrate"),
stratum_col = "stratum"
)Arguments
- design
A
creel_designobject with at least one survey sub-object already attached (add_countsoradd_interviews).- response_rates
A data frame or tibble with one row per stratum, containing at minimum the columns:
- stratum
Character or factor. Stratum identifier matching the values in the design's strata column.
- n_sampled
Integer. Number of units approached for interview in the stratum.
- n_responded
Integer. Number of units that actually responded. Must be
<= n_sampled.
- method
Character. Weighting method to apply. Either
"postStratify"(default, adjusts weights to match known stratum totals inversely proportional to response rate) or"calibrate"(calibration estimator viasurvey::calibrate; requires stratum population totals inresponse_rates).- stratum_col
Character. Name of the stratum column in
response_rates(default:"stratum").
Value
The input creel_design with adjusted weights. The updated
design includes an attribute "nonresponse_diagnostics" (a tibble
with columns stratum, n_sampled, n_responded,
response_rate, weight_adjustment) that can be retrieved
with attr(result, "nonresponse_diagnostics").
Adjustment method
For each stratum \(h\): $$response\_rate_h = n\_responded_h / n\_sampled_h$$ $$weight\_adjustment_h = 1 / response\_rate_h$$
The original weights are multiplied by weight_adjustment_h, which
upweights respondents to represent non-respondents (Armstrong & Overton 1977).
This assumes that respondents and non-respondents are exchangeable within
strata (a missing-at-random assumption). When this is implausible, a
sensitivity analysis comparing pre- and post-adjustment estimates is
recommended.
References
Armstrong, B.G. and Overton, W.S. 1977. Estimating nonresponse bias in mail surveys. Journal of Marketing Research 14:396–402.
Pollock, K.H., Jones, C.M. and Brown, T.L. 1994. Angler Survey Methods and Their Applications in Fisheries Management. American Fisheries Society, Bethesda, MD.
See also
Other "Reporting & Diagnostics":
check_completeness(),
compare_variance(),
flag_outliers(),
season_summary(),
standardize_species(),
summarize_by_angler_type(),
summarize_by_day_type(),
summarize_by_method(),
summarize_by_species_sought(),
summarize_by_trip_length(),
summarize_cws_rates(),
summarize_hws_rates(),
summarize_length_freq(),
summarize_refusals(),
summarize_successful_parties(),
summarize_trips(),
summary.creel_estimates(),
validate_creel_data(),
validate_design(),
validate_incomplete_trips(),
validation_report(),
write_estimates()
Examples
data("example_counts", package = "tidycreel")
data("example_interviews", package = "tidycreel")
cal <- unique(example_counts[, c("date", "day_type")])
design <- creel_design(cal, date = date, strata = day_type)
design <- suppressWarnings(add_counts(design, example_counts))
design <- suppressWarnings(add_interviews(
design, example_interviews,
catch = catch_total, effort = hours_fished,
trip_status = trip_status, trip_duration = trip_duration
))
#> ℹ No `n_anglers` provided — assuming 1 angler per interview.
#> ℹ Pass `n_anglers = <column>` to use actual party sizes for angler-hour
#> normalization.
#> ℹ Added 22 interviews: 17 complete (77%), 5 incomplete (23%)
resp <- data.frame(
stratum = c("weekday", "weekend"),
n_sampled = c(80L, 60L),
n_responded = c(72L, 48L)
)
adj_design <- adjust_nonresponse(design, resp)
attr(adj_design, "nonresponse_diagnostics")
#> # A tibble: 2 × 5
#> stratum n_sampled n_responded response_rate weight_adjustment
#> <chr> <int> <int> <dbl> <dbl>
#> 1 weekday 80 72 0.9 1.11
#> 2 weekend 60 48 0.8 1.25