順序のあるカテゴリカルデータ分割表解析

  • Joint Statistical Meeting 2010 @ Vancouverの講義(2010/08/02)(主なセッションはこちら)
  • Analysis of Ordinal Categorical Data
  • 種本

Analysis of Ordinal Categorical Data (Wiley Series in Probability and Statistics)

Analysis of Ordinal Categorical Data (Wiley Series in Probability and Statistics)

    • Contents
      • 1. Logistic Regression Models Using Cumulative Logits ("Proportional Odds" and extensions)
      • 2. Other Ordinal Logistic Modeling (adjacent-categories and continuation-ratio logits, sterotype model, case-control studies, Bayesian approach)
      • 3. Other Ordinal Multinomial Models (cumulative probit, log-log links, mean sresponse model)
      • 4. Clustered Ordinal Responses: Marginal Models and Random Effects Models (cumulative logit marginal GEE and random Effects Models)
      • 5. Software summary
      • 6. Summarizing other research work on ordinal modeling included for your reference but not covered in the lectures
  • N\times MテーブルにおけOdds ratioのいろいろ
    • Local odds ratio
      • 隣接する2\times 2部分テーブルのOR
      • \frac{n_{i,j}n_{i+1,j+1}}{n_{i,j+1}n_{i+1,j}}
    • Global odds ratio
      • ある行・列を境に、テーブル全体を2\times 2に分割し、そのOR
      • \frac{(\sum_{a\le i} \sum_{b\le j} n_{a,b})(\sum_{a > i} \sum_{b > j} n_{a,b})}{(\sum_{a\le i} \sum_{b> j} n_{a,b})(\sum_{a> i} \sum_{b\le j} n_{a,b})}
    • Cumulative odds ratio
      • ある隣接2行について、ある列を境に(細長い)2\times 2テーブルを作り、そのOR
        • \frac{( \sum_{b \le j} n_{i,b})( \sum_{b > j} n_{i+1,b})}{( \sum_{b> j} n_{i,b})( \sum_{b \le j} n_{i+1,b})}
    • Continuation odds ratio
      • ある隣接2行について、ある列を境に、特定の列と、それより右側の全列とで、2\times 2テーブル(そのうち2セルは、1セルからなり、残りの2セルは、細長い1行1列セルからなる)のOR
      • \frac{(n_{i,j})( \sum_{b > j} n_{i+1,b})}{( \sum_{b> j} n_{i,b})( n_{i+1,j})}
  • Ordinal Logistic Regression Models
    • Adjacent-category logits (ACL)
      • \log (\frac{P(y_i=j)}{P(y_i=j+1)})=\alpha_j + \beta' x_i
    • Baseline-category logits (BCL)
      • \log (\frac{\pi_j}{\pi_c})=\sum_{k=j+1}^c \log(\frac{\pi_{k-1}}{\pi_{k}})
    • Continuation-ratio logits
      • \log (\frac{\pi_j}{\sum_{k=j+1}^c \pi_{k}}) = \log (\frac{w_j}{(1-w_j)}
    • Stereotype model: Multiplicative paired-category logits
      • \log(\frac{\pi_j}{\pi_c}) = \alpha_j +\phi \beta' x
        • Constraint 1
          • Discrete
          • \phi = 0 \text{or} 1
        • Constraint 2
          • Continuous
          • \phi_1 \ge \phi_2 \ge ... \ge \phi_c or \phi_1 \le \phi_2 \le ... \le \phi_c
    • Proportional Odds ならRでpolr()によるordered logistic regression
  • その他
    • generalized estimating equation (GEE)
    • Generalized linear mixed model (GLMM)
  • Missing dataの分類(単語化は力)
    • Missing completely at random(MCAR)
    • Missing at random(MAR)
    • Not missing at random(NMAR)
  • Missing dataへの対応(単語化は力)
    • Complete case analysis(Missing dataのないサンプルのみで解析)
    • Imputation
    • Likelihood methods
    • Weighting inversely by probability of response