Required:
  - main_analysis: "Analysis method used"
  - main_results: "Effect sizes, p-values, etc."
  - analytical_decisions: "Choices made"

Optional:
  - alternative_choices: "Alternatives considered"
  - concerns: "Specific aspects to verify"

## Sensitivity Analysis Plan

### 1. Analytical Decision Inventory

| Decision Area | Main Analysis Choice | Alternative 1 | Alternative 2 | Alternative 3 |
|--------------|---------------------|---------------|---------------|---------------|
| Outlier handling | Exclude 3SD | Exclude 2SD | Include | Winsorize |
| Missing data | Listwise | Pairwise | MI | FIML |
| Control variables | A, B, C | A, B | A, B, C, D | None |
| Statistical model | OLS | Robust SE | Bootstrap | MLM |
| Sample restriction | All | Condition1 only | Condition2 only | |

**Total specification count**: [N] (= 4 × 4 × 4 × 4 × 3)

### 2. Sensitivity Analysis Plan

#### A. Outlier Analysis

**Purpose**: Evaluate impact of extreme values on results

**Methods**:
1. Identify influential cases using Cook's D criterion (D > 4/n)
2. Reanalyze after excluding influential cases
3. Apply various outlier criteria (2SD, 3SD, IQR)

**Expected Results**:
| Condition | Effect Size | p-value | Conclusion Consistency |
|-----------|-------------|---------|----------------------|
| Main analysis | [d] | [p] | - |
| Exclude Cook's D | | | Yes/No |
| Exclude 2SD | | | Yes/No |
| Exclude IQR | | | Yes/No |

#### B. Missing Data Handling Analysis

**Purpose**: Evaluate impact of missing data handling methods on results

**Methods**:
1. Listwise deletion (main analysis)
2. Pairwise deletion
3. Multiple imputation (m=20)
4. Full Information Maximum Likelihood (FIML)

**Expected Results**:
| Method | N | Effect Size | 95% CI | p-value |
|--------|---|-------------|--------|---------|
| Listwise | | | | |
| Pairwise | | | | |
| MI (m=20) | | | | |
| FIML | | | | |

#### C. Control Variable Combination Analysis

**Purpose**: Evaluate impact of control variable selection on results

**Combinations**:
1. No control variables (bivariate)
2. Core control variables only (A, B)
3. All control variables (A, B, C) - main analysis
4. Extended control variables (A, B, C, D)

**Expected Results**:
| Model | Control Variables | β | SE | p |
|-------|------------------|---|----|----|
| Model 0 | None | | | |
| Model 1 | A, B | | | |
| Model 2 | A, B, C | | | |
| Model 3 | A, B, C, D | | | |

#### D. Alternative Statistical Models

**Purpose**: Evaluate impact of model specification changes

**Alternative Models**:
1. OLS with HC robust SE
2. Bootstrap (1000 iterations)
3. Bayesian regression
4. Quantile regression (median)

### 3. Specification Curve Analysis

**Analysis Specification Elements**:

**Visualization Plan**:

  Effect Size Distribution
  ↑
  │    ●●●●●●●●●●●●●●●●●●●●●●●●●●
  │  ●●                          ●●
  │●                                ●

**Result Interpretation Criteria**:
- Robust: Same direction + significant in XX% or more specifications
- Partially robust: Same direction in XX% or more (regardless of significance)
- Unstable: Same direction in less than XX%

### 4. Leave-One-Out Analysis (for meta-analysis)

**Purpose**: Evaluate impact of individual studies on overall effect

**Results Table**:
| Excluded Study | k | Effect Size | 95% CI | Change |
|----------------|---|-------------|--------|--------|
| (None) | [N] | [d] | [CI] | - |
| Study 1 | N-1 | | | |
| Study 2 | N-1 | | | |
| ... | | | | |

### 5. Results Synthesis and Interpretation

**Robustness Evaluation Criteria**:
- ✅ Robust: Main conclusion maintained across all sensitivity analyses
- ⚠️ Conditionally robust: Conclusion maintained only under some conditions
- ❌ Unstable: Conclusion sensitive to analytical decisions

**Final Evaluation**: [Evaluation result]

**Reporting Summary**:
"The main analysis result (d = X.XX, p = .XXX) was consistently observed
in [M] out of [N] alternative analysis specifications (XX%).
In particular, results were [stable/sensitive] to changes in [decision]."

You are a sensitivity analysis expert.

Please design a strategy to verify the robustness of the following analysis results:

[Main Analysis]: {main_analysis}
[Main Results]: {main_results}
[Analytical Decisions]: {analytical_decisions}

Tasks to perform:
1. List analytical decisions
   - Data preprocessing decisions
   - Inclusion/exclusion criteria
   - Statistical model selection
   - Control variable selection
   - Outlier handling

2. Alternative specifications for each decision
   | Decision | Main Analysis Choice | Alternative 1 | Alternative 2 |

3. Sensitivity analysis plan
   - Leave-one-out analysis
   - Alternative model specifications
   - Alternative missing data handling
   - Alternative outlier criteria

4. Specification Curve analysis
   - All reasonable analysis specification combinations
   - Result distribution visualization

5. Result interpretation criteria
   - Robust conclusion: Same direction in XX% or more specifications
   - Unstable conclusion: Same direction in less than XX%

6. Reporting format
   - Sensitivity analysis results table
   - Specification curve graph

library(specr)

# Setup specifications
specs <- setup(
  data = data,
  y = c("dv1", "dv2"),           # DV options
  x = c("iv1", "iv2"),           # IV options
  model = c("lm", "lm_robust"),  # Model options
  controls = c("c1", "c1 + c2")  # Control variable options
)

# Run analysis
results <- specr(specs)

# Visualization
plot(results)

library(metafor)

# Leave-one-out analysis
loo <- leave1out(rma_model)

# Visualization
forest(loo)