--- title: "Confidence Boosting and Sensitivity Analysis" output: rmarkdown::html_vignette vignette: > %\VignetteIndexEntry{Confidence Boosting and Sensitivity Analysis} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- ```{r, setup, include = FALSE} knitr::opts_chunk$set( collapse = TRUE, comment = "#>" ) ``` ```{r load-and-setup, echo=FALSE, message=FALSE, warning=FALSE} library(andorR) data(ethical) ``` ## Introduction: From Conclusion to Conviction In any complex analysis, the first pieces of evidence are often the easiest to acquire. They may come from readily available data, quick observations, or even an expert's "gut feeling." The `andorR` **Discovery Phase**, guided by the `influence_index`, is designed to use this initial, often low-confidence, information to reach a preliminary conclusion as quickly as possible. However, once a conclusion is reached (e.g., the root is `TRUE` or `FALSE`), the job is often not finished. The overall confidence in that conclusion may be too low for a final report or a high-stakes decision. This is where the **Confidence Boosting Phase** begins. Instead of investing resources to improve the evidence for *every* question, `andorR` allows you to strategically identify which actions—either answering a new question or improving an existing answer—will most efficiently increase your confidence in the final result. ## The Algorithm: A Sensitivity Analysis The `get_confidence_boosters()` function is the engine for this phase. It performs a **sensitivity analysis** on the current state of the tree to find the most effective next steps. It does this in two ways: 1. **Analyzing Unanswered Questions:** For every remaining leaf with an `NA` answer, the function simulates answering it with 100% confidence as `TRUE`, and then again as `FALSE`. It calculates the potential increase in the root's confidence for both scenarios and records the more beneficial of the two. 2. **Analyzing Existing Answers:** For every leaf that has already been answered with less than 100% confidence (i.e., a score of 0-4), the function simulates what would happen if you invested the effort (e.g., through more research) to increase its confidence to the maximum (5/5). It then calculates the potential gain. Finally, the function combines all these potential gains into a single, ranked list, showing you the actions that give you the most "bang for your buck" in terms of increasing your final confidence. ## A Worked Example: Strengthening a Conclusion Let's start with a scenario where we have reached a preliminary conclusion for the `ethical` investment tree, but our confidence is not yet high enough. ### Step 1: Reach an Initial Conclusion First, we load the tree and provide several answers, some with high and some with low confidence, to simulate an initial assessment. ```{r setup-tree-boosting} library(andorR) data(ethical) dtree <- load_tree_df(ethical) # Provide some initial answers with varying confidence set_answer(dtree, "FIN1", TRUE, 5) # Parent "Profitability" becomes TRUE set_answer(dtree, "FIN4", TRUE, 2) # Low confidence set_answer(dtree, "FIN5", TRUE, 2) # Low confidence set_answer(dtree, "ENV1", TRUE, 4) set_answer(dtree, "ENV2", TRUE, 3) set_answer(dtree, "ENV3", TRUE, 4) # Parent "Clean Record" becomes TRUE set_answer(dtree, "SOC2", TRUE, 4) set_answer(dtree, "GOV1", TRUE, 5) set_answer(dtree, "GOV2", TRUE, 5) set_answer(dtree, "GOV3", TRUE, 5) set_answer(dtree, "GOV4", TRUE, 5) set_answer(dtree, "GOV5", TRUE, 5) # Parent "Strong Corporate Governance" becomes TRUE # Tree root becomes TRUE # Recalculate the tree state dtree <- update_tree(dtree) ``` ### Step 2: View the Initial Conclusion Let's print the tree. We can see that we have reached a `TRUE` conclusion, but the confidence is only **28.6%**, which is likely to be too low. ```{r print-initial-conclusion} print_tree(dtree) ``` ### Step 3: Get Guidance on Boosting Confidence Now we enter the Confidence Boosting phase. We call `get_confidence_boosters()` to get a ranked list of the most effective actions to take next. ```{r get-guidance} guidance <- get_confidence_boosters(dtree) knitr::kable(guidance, caption = "Top Actions to Boost Confidence") ``` The guidance table clearly shows that the most effective action we can take is to **Answer new questions** in our tree for the **ENV** questions (ENV4 to ENV6) or **Increase confidence** in FIN4 or FIN5. ### Step 4: Act on the Guidance Let's follow the top suggestion. We'll "do more research" on `FIN5` and update its confidence to the maximum level of 5. ```{r act-on-guidance} # Edit the existing answer for FIN5 with a new, higher confidence set_answer(dtree, "FIN5", TRUE, 5) # Recalculate the entire tree dtree <- update_tree(dtree) ``` ### Step 5: View the Improved Result Let's print the tree again. As predicted by the guidance, our final confidence in the `TRUE` result has jumped from 28.6% to **40.8%**. ```{r final-print} print_tree(dtree) ``` Iteratively recalculating the tree, and regenerating the list of the best questions to boost confidence will progressively guide the most efficient way to increase confidence. This two-stage iterative workflow is implemented in the `andorR_interactive()` function. This approach allows you to move quickly to a conclusion with readily available information, and then strategically invest your resources to strengthen that conclusion to the desired level of certainty.