Literature Gap Finder

Method-Setting Matrix

Systematic Gap Identification Framework

The method-setting matrix is the core tool for finding research gaps systematically:

# Build a method-setting matrix programmatically
create_gap_matrix <- function() {
  methods <- c("Regression", "Weighting/IPW", "DR/AIPW", "TMLE", "ML-based")
  settings <- c("Binary treatment", "Continuous treatment",
                "Time-varying", "Clustered", "High-dimensional",
                "Measurement error", "Missing data", "Network")

  matrix_data <- expand.grid(method = methods, setting = settings)
  matrix_data$status <- "unknown"  # To be filled: "developed", "partial", "gap"
  matrix_data$priority <- NA
  matrix_data$references <- ""

  matrix_data
}

# Visualize the gap matrix
visualize_gaps <- function(gap_matrix) {
  library(ggplot2)

  ggplot(gap_matrix, aes(x = method, y = setting, fill = status)) +
    geom_tile(color = "white") +
    scale_fill_manual(values = c(
      "developed" = "#2ecc71",
      "partial" = "#f39c12",
      "gap" = "#e74c3c",
      "unknown" = "#95a5a6"
    )) +
    theme_minimal() +
    labs(title = "Method × Setting Gap Matrix",
         x = "Method", y = "Setting") +
    theme(axis.text.x = element_text(angle = 45, hjust = 1))
}

Verification Process

Confirming a Gap is Real

Before claiming a gap, verify systematically:

# Systematic verification checklist
verify_gap <- function(topic, keywords) {
  checklist <- list(
    databases_searched = c("google_scholar", "web_of_science", "pubmed", "scopus"),
    search_terms = keywords,
    date_range = paste(Sys.Date() - 365*5, "to", Sys.Date()),
    results = list(
      papers_found = 0,
      closest_related = c(),
      why_not_the_same = ""
    ),
    expert_consultation = list(
      experts_contacted = c(),
      responses = c()
    ),
    verification_status = "pending"  # pending, confirmed, rejected
  )

  checklist
}

# Document the verification
document_verification <- function(gap_description, search_log) {
  cat("## Gap Verification Report\n\n")
  cat("**Gap:**", gap_description, "\n\n")
  cat("**Search Date:**", as.character(Sys.Date()), "\n\n")
  cat("**Databases Searched:**\n")
  for (db in search_log$databases_searched) {
    cat("- ", db, "\n")
  }
  cat("\n**Search Terms:**", paste(search_log$search_terms, collapse = ", "), "\n")
  cat("\n**Conclusion:**", search_log$verification_status, "\n")
}

Priority Ranking

Scoring Research Gaps

Priority Score = Σ(weight × score)

# Priority scoring function
score_research_gap <- function(
  impact,        # 1-5: How many researchers would benefit
  novelty,       # 1-5: How new/original is this
  tractability,  # 1-5: How likely can we solve it
  timeliness,    # 1-5: Is this currently hot
  fit,           # 1-5: Matches our expertise
  publication    # 1-5: Publication potential
) {
  weights <- c(0.25, 0.20, 0.20, 0.15, 0.10, 0.10)
  scores <- c(impact, novelty, tractability, timeliness, fit, publication)

  priority <- sum(weights * scores)

  list(
    priority_score = priority,
    interpretation = case_when(
      priority >= 4.0 ~ "High priority - pursue immediately",
      priority >= 3.0 ~ "Medium priority - develop further",
      priority >= 2.0 ~ "Low priority - back burner",
      TRUE ~ "Skip - not worth pursuing"
    ),
    breakdown = data.frame(
      criterion = c("Impact", "Novelty", "Tractability",
                   "Timeliness", "Fit", "Publication"),
      weight = weights,
      score = scores,
      weighted = weights * scores
    )
  )
}

# Compare multiple gaps
rank_gaps <- function(gaps_list) {
  scores <- sapply(gaps_list, function(g) g$priority_score)
  order(scores, decreasing = TRUE)
}

Method × Setting Matrix

The Core Framework

Systematically map methods against settings to find gaps:

                    METHODS
          │ Regression │ Weighting │ DR/TMLE │ ML-based │
──────────┼────────────┼───────────┼─────────┼──────────│
Binary A  │     ✓      │     ✓     │    ✓    │    ✓     │
Continuous│     ✓      │     ?     │    ✓    │    ?     │
SETTINGS  ├────────────┼───────────┼─────────┼──────────│
Time-vary │     ?      │     ✓     │    ✓    │    ✗     │
Clustered │     ✓      │     ?     │    ?    │    ✗     │
High-dim  │     ✗      │     ✗     │    ?    │    ✓     │

✓ = Well-developed    ? = Partial/emerging    ✗ = Gap

Building Your Matrix

Step 1: Identify Dimensions

For mediation analysis:

Step 2: List Methods

Step 3: Fill and Analyze

Mark each cell:

• ✓ (green): Well-established with theory + software
• ? (yellow): Emerging or partial coverage
• ✗ (red): Clear gap

Example: Sequential Mediation Matrix

                         │ Product │ Weighting │ DR │ Bounds │
─────────────────────────┼─────────┼───────────┼────┼────────│
2 mediators, linear      │    ✓    │     ✓     │  ✓ │   ?    │
2 mediators, nonlinear   │    ?    │     ✓     │  ? │   ✗    │
3+ mediators, linear     │    ?    │     ?     │  ✗ │   ✗    │
3+ mediators, nonlinear  │    ✗    │     ?     │  ✗ │   ✗    │
With measurement error   │    ✗    │     ✗     │  ✗ │   ✗    │
With unmeasured conf.    │    ✗    │     ✗     │  ✗ │   ?    │

Gaps identified:

• DR methods for 3+ mediators
• Any method with measurement error
• Bounds approach underdeveloped

Assumption Relaxation Trees

The Framework

Map how assumptions have been relaxed over time:

                    Standard Mediation (Baron-Kenny 1986)
                              │
            ┌─────────────────┼─────────────────┐
            ↓                 ↓                 ↓
    No unmeasured      Linearity         No interaction
    confounding        assumed           assumed
            │                 │                 │
            ↓                 ↓                 ↓
    ┌───────┴───────┐   Nonparametric    VanderWeele
    ↓               ↓     (Imai 2010)    4-way decomp
Sensitivity      Bounds                        │
(Imai 2010)   (partial ID)                    ↓
    │               │               Multiple mediators?
    ↓               ↓               Longitudinal?
 E-value        Sharp bounds?       Measurement error?
(Ding 2016)         │                    │
    │               ↓                    ↓
    ↓           [YOUR GAP?]         [YOUR GAP?]
[YOUR GAP?]

Building the Tree

Step 1: Identify Original Assumptions

For a classic method, list ALL assumptions:

1. Explicit assumptions (stated in paper)
2. Implicit assumptions (unstated but required)
3. Computational assumptions (required for implementation)

Step 2: Trace Relaxation History

For each assumption, find papers that:

• Relax it partially
• Relax it completely
• Replace it with different assumption
• Show consequences of violation

Step 3: Find Unexplored Branches

Look for:

• Combinations of relaxations not yet explored
• Relaxations in one method not applied to another
• Partial relaxations that could be completed

Example: Positivity Assumption

Positivity: P(A=a|X) > ε > 0 for all a, x
                    │
    ┌───────────────┼───────────────┐
    ↓               ↓               ↓
Near-violation  Practical      Structural
                positivity      violations
    │               │               │
    ↓               ↓               ↓
Trimming      Overlap         Extrapolation
weights       assessment       methods
    │               │               │
    ↓               ↓               ↓
Truncation?   Diagnostics?   Bounds under
                             violations?

Citation Network Analysis

Forward and Backward Searching

Backward: From recent key paper, trace citations:

• What foundational papers are cited?
• What parallel developments exist?
• What's the intellectual lineage?

Forward: Using Google Scholar "Cited by":

• Who has built on this work?
• What extensions were made?
• What gaps remain unaddressed?

Key Paper Identification

For any topic, identify:

Building a Citation Map