---
title: "Cookbook: Proportion Outcome, End to End"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Cookbook: Proportion Outcome, End to End}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r, include = FALSE}
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
```

One complete, runnable script for a **proportion** outcome — a response
that is itself a fraction in (0, 1) per subject (adherence rate, fraction
of tissue affected, score normalized to a 0–1 scale). This is distinct
from the incidence cookbook, where each subject contributes a single 0/1.
Same four steps as every cookbook (see `vignette("cookbook-continuous")`);
the inference classes are the `InferenceProp*` family — fractional logit
(quasi-binomial), Beta regression, and zero/one-inflated Beta for data
with exact 0s and 1s.

## Setup

EDI is not on CRAN yet, so `install.packages("EDI")` fails — install from
R-universe (fallback: GitHub, `subdir = "R/EDI"`). Not evaluated here.

```{r install, eval = FALSE, purl = FALSE}
install.packages("EDI", repos = c("https://kapelner.r-universe.dev", "https://cloud.r-project.org"))
# or: remotes::install_github("kapelner/EDI", subdir = "R/EDI")
```

```{r setup}
library(EDI)
set.seed(20260916)

n = 80
X = data.frame(
  severity = round(runif(n, 1, 10), 1),
  prior    = rbinom(n, 1, 0.4)
)
true_logit_shift = 0.7
```

## Fixed design, fractional logit

The response is generated from a Beta distribution whose mean follows a
logistic model in treatment and covariates, then kept strictly inside
(0, 1) — fractional logit and Beta regression require open-interval data;
the zero/one-inflated class handles exact boundary values.

```{r fixed}
des = DesignFixedBernoulli$new(n = n, response_type = "proportion", verbose = FALSE)
des$add_all_subjects_to_experiment(X)
des$assign_w_to_all_subjects()
w = des$get_w()

mu  = plogis(-0.5 + true_logit_shift * w - 0.1 * X$severity + 0.4 * X$prior)
phi = 15                                  # Beta precision
y   = rbeta(n, mu * phi, (1 - mu) * phi)
y   = pmin(pmax(y, 1e-4), 1 - 1e-4)       # keep strictly inside (0, 1)
des$add_all_subject_responses(y)

inf = InferencePropFractionalLogit$new(des, verbose = FALSE)
inf$num_cores = 1L
inf$compute_estimate()                    # treatment effect on the logit scale
inf$compute_asymp_confidence_interval(alpha = 0.05)
inf$compute_asymp_two_sided_pval()
```

```{r fixed-resampling}
inf$set_seed(1)
inf$compute_rand_two_sided_pval(r = 200, show_progress = FALSE)
inf$set_seed(1)
inf$compute_bootstrap_confidence_interval(alpha = 0.05, B = 200, show_progress = FALSE)
```

## Beta regression on the same design

```{r beta}
inf_beta = InferencePropBetaRegr$new(des, verbose = FALSE)
inf_beta$num_cores = 1L
inf_beta$compute_estimate()
inf_beta$compute_asymp_confidence_interval(alpha = 0.05)
```

## Everything at once

```{r suite}
suite = InferenceSuite$new(des)
res = suite$run_all_inference(screen = TRUE, plots = FALSE, num_cores = 1L,
                              methods = c("wald", "score", "lik_ratio"), max_secs_per_class = 15)
```

## Sequential design

```{r seq}
des_seq = DesignSeqOneByOneKK14$new(n = n, response_type = "proportion", verbose = FALSE)
for (i in seq_len(n)) {
  w_i  = des_seq$add_one_subject_to_experiment_and_assign(X[i, , drop = FALSE])
  mu_i = plogis(-0.5 + true_logit_shift * w_i - 0.1 * X$severity[i] + 0.4 * X$prior[i])
  y_i  = rbeta(1, mu_i * phi, (1 - mu_i) * phi)
  des_seq$add_one_subject_response(i, min(max(y_i, 1e-4), 1 - 1e-4))
}

inf_seq = InferencePropFractionalLogit$new(des_seq, verbose = FALSE)
inf_seq$num_cores = 1L
inf_seq$compute_estimate()
inf_seq$set_seed(1)
inf_seq$compute_rand_two_sided_pval(r = 200, show_progress = FALSE)
```

## Where to go next

- Exact 0s and 1s in the data: `InferencePropZeroOneInflatedBetaRegr`.
- A marginal mean difference on the original 0–1 scale rather than a
  logit coefficient: `InferencePropGCompMeanDiff`.
- `vignette("validation-evidence")`: checked against `betareg::betareg`
  and `stats::glm(family = quasibinomial)`.