Skip to contents

Internal Estimation Function

Source: R/pj_estimate.R
pj_estimate.Rd

Core workhorse for computing marginal means (MMs) or AMCEs from a conjoint design, with optional intra-respondent reliability (IRR) correction.

Usage

pj_estimate(
  .data,
  .structure,
  .estimand,
  .att_choose,
  .lev_choose,
  .att_notchoose,
  .lev_notchoose,
  .att_choose_b,
  .lev_choose_b,
  .att_notchoose_b,
  .lev_notchoose_b,
  .se_method,
  .irr,
  .remove_ties,
  .ignore_position,
  .n_sims,
  .n_boot,
  .weights_1,
  .clusters_1,
  .se_type_1,
  .weights_2,
  .clusters_2,
  .se_type_2,
  .auto_cluster = TRUE,
  .seed = NULL
)

Arguments

.data

A projoint_data object created by reshape_projoint or make_projoint_data.

.structure

Either "profile_level" or "choice_level".

.estimand

Either "mm" (marginal mean) or "amce" (average marginal component effect).

.att_choose

Attribute for the chosen profile/feature.

.lev_choose

Level(s) for the chosen profile/feature. Length 1 for profile_level; may be 1+ for choice_level.

.att_notchoose

Attribute for the not-chosen profile/feature (required for choice_level).

.lev_notchoose

Level(s) for the not-chosen profile/feature (required for choice_level).

.att_choose_b

(AMCE only) Baseline attribute for comparison.

.lev_choose_b

(AMCE only) Baseline level(s) for comparison.

.att_notchoose_b

(AMCE only, choice-level only) Baseline attribute for the not-chosen profile.

.lev_notchoose_b

(AMCE only, choice-level only) Baseline level(s) for the not-chosen profile.

.se_method

One of "analytical", "simulation", or "bootstrap".

.irr

NULL (default) to estimate IRR from repeated tasks; otherwise a numeric IRR value.

.remove_ties

Logical; should ties be removed before estimation? Defaults to TRUE.

.ignore_position

Logical; only for choice_level. If TRUE (default), ignore profile position (left/right).

.n_sims

Integer; required if .se_method == "simulation".

.n_boot

Integer; required if .se_method == "bootstrap".

.weights_1

(Optional) Bare (unquoted) column with weights for IRR estimation; passed to lm_robust.

.clusters_1

(Optional) Bare (unquoted) column with clusters for IRR estimation; passed to lm_robust.

.se_type_1

SE type for IRR estimation; passed to lm_robust. If NULL, estimatr defaults are used (HC2 when unclustered; CR2 when clustered).

.weights_2

(Optional) Bare (unquoted) column with weights for MM/AMCE estimation; passed to lm_robust.

.clusters_2

(Optional) Bare (unquoted) column with clusters for MM/AMCE estimation; passed to lm_robust.

.se_type_2

SE type for MM/AMCE estimation; passed to lm_robust. If NULL, estimatr defaults are used (HC2 when unclustered; CR2 when clustered).

.auto_cluster

Logical; if TRUE (default), auto-cluster on id when present and no .clusters_* are supplied; auto-clustering only occurs when the corresponding .se_type_* is NULL. See projoint.

.seed

Optional integer. If supplied, sets a temporary RNG seed for reproducible simulation/bootstrap inside this call and restores the previous RNG state on exit.

Value

A data frame with rows for the requested estimand(s) and columns:

  • estimand: one of "mm_uncorrected", "mm_corrected", "amce_uncorrected", "amce_corrected".

  • estimate, se, conf.low, conf.high, tau.

Details

IRR is clipped to [0.5, 1) (with a tiny epsilon) to avoid boundary issues. For choice-level MMs, ties must be removed (.remove_ties = TRUE). When .seed is supplied, the previous RNG state is restored on exit.

See also

lm_robust, projoint, projoint_level, projoint_diff