---
title: "Comparing probmed Workflows"
format: html
vignette: >
  %\VignetteIndexEntry{Comparing probmed Workflows}
  %\VignetteEngine{quarto::html}
  %\VignetteEncoding{UTF-8}
---

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

The `probmed` package is designed to be flexible, allowing you to extract mediation structures from various R objects. This vignette compares the different workflows available:

1.  **Base R (`lm`/`glm`)**: Best for simple models and maximum control.
2.  **`lavaan`**: Best for Structural Equation Models (SEM), latent variables, and complex path structures.
3.  **`mediation`**: Best if you are already using the `mediation` package for other analyses (e.g., sensitivity analysis).

## Feature Comparison

| Feature | `lm` / `glm` | `lavaan` | `mediation` |
| :--- | :---: | :---: | :---: |
| **Model Types** | Linear, GLM | SEM, CFA, Path Analysis | Linear, GLM, GAM, etc. |
| **Latent Variables** | No | Yes | No |
| **Multiple Mediators** | Manual setup | Native support | Native support |
| **Missing Data** | `na.action` | FIML (`missing="fiml"`) | `na.action` |
| **Standardized Est.** | Manual | Native (`standardized=TRUE`) | Manual |
| **Ease of Use** | High | Medium | High |

## Workflow 1: Base R (`lm`/`glm`)

This is the most direct method. You fit two separate models: one for the mediator and one for the outcome.

```{r}
#| eval: false
library(probmed)

# Simulate data
set.seed(123)
n <- 100
data <- data.frame(X = rnorm(n))
data$M <- 0.5 * data$X + rnorm(n)
data$Y <- 0.4 * data$M + 0.2 * data$X + rnorm(n)

# Fit models
model_m <- lm(M ~ X, data = data)
model_y <- lm(Y ~ M + X, data = data)

# Extract and compute
extract <- extract_mediation(model_m, model_y = model_y, treatment = "X", mediator = "M")
pmed(extract)
```

**Pros**:
*   No extra dependencies.
*   Full control over model specification.

**Cons**:
*   Requires managing two model objects.
*   Manual handling of missing data consistency.

## Workflow 2: `lavaan`

Ideal for users familiar with SEM or when dealing with latent variables or FIML.

```{r}
#| eval: false
library(lavaan)

model <- '
  M ~ a*X
  Y ~ b*M + cp*X
'

fit <- sem(model, data = data)

extract <- extract_mediation(fit, treatment = "X", mediator = "M")
pmed(extract)
```

**Pros**:
*   Single model specification.
*   Handles FIML for missing data.
*   Supports latent variables (though `probmed` currently requires observed data for bootstrapping).

**Cons**:
*   Requires `lavaan` package.
*   Syntax is different from standard R formulas.

## Workflow 3: `mediation`

If you use the `mediation` package for its specific features (like sensitivity analysis), you can easily pipe the results to `probmed`.

```{r}
#| eval: false
library(mediation)

med_out <- mediate(model_m, model_y, treat = "X", mediator = "M")

extract <- extract_mediation(med_out)
pmed(extract)
```

**Pros**:
*   Seamless integration with `mediation` workflow.
*   Automatic variable detection.

**Cons**:
*   Requires `mediation` package.
*   Limited to models supported by `mediate()`.

## Conclusion

Choose the workflow that best fits your existing analysis pipeline. `probmed` ensures that regardless of the input method, the calculation of $P_{med}$ remains consistent and robust.