关于缺失GOS数据处理：listwise与pairwise deletion的区分咨询

Hey there, let's walk through your questions step by step—since you're zeroing in on missing GOS values and trying to clarify listwise vs pairwise deletion, I’ll break down the core differences first, then apply them to the study designs you’ve observed.

• Listwise Deletion (Complete Case Analysis)
  This method removes an entire study participant from all analyses if they have a missing value for any variable you’re working with (whether that’s a predictor or the GOS outcome). Put simply: only participants with complete data across every variable in your dataset get included in any statistical work.

• Pairwise Deletion
  This approach is more flexible: for each specific analysis you run, you only exclude participants who have missing values for the specific variables involved in that analysis. Different analyses might use slightly different subsets of your sample. For example:

  • When calculating correlations between two predictors, you’d only exclude people missing either of those two predictors.
  • When analyzing the relationship between a predictor and GOS, you’d only exclude people missing that predictor or GOS.
    The key here is that no participant is fully excluded from all analyses—they’re only left out of the specific tests where their data is incomplete.

Let’s map the two scenarios you described to these methods:

1. Scenario 1: Studies report predictors for all enrolled patients, but only use those with GOS data for outcome analysis
   This does not count as pairwise deletion. Here’s why:

• The researchers are using two distinct sample sets: all enrolled patients for describing predictors, and only those with complete GOS data for outcome-related analyses.
• For the outcome analysis itself, they’re using a complete case subset (only participants with non-missing GOS), but this isn’t pairwise deletion because pairwise requires using variable-specific subsets across multiple analyses of the same core relationships. This is better described as using a complete case sample for outcome analysis while retaining the full cohort for descriptive predictor reporting.

2. Scenario 2: Studies set GOS completeness as an inclusion criterion, and don’t report predictors for patients with missing GOS
   You’re exactly right—this is classic listwise deletion. By making GOS completeness a requirement for study entry, the researchers are excluding anyone with missing GOS from the entire study from the start. All analyses (predictor descriptions, outcome analyses, etc.) are done on a cohort where every participant has complete GOS data, which fits the definition of listwise deletion perfectly.

内容的提问来源于stack exchange，提问作者user193843