統計資料處理之原則

Principles of Statistical Data Handling

本文旨在說明從事資料處理時，須注意之原則，以免造成「垃圾進，垃圾出」之結果。本文以SAS統計軟體為例，內容包括：一、統計資料處理應注意事項：「原子」原則、適當性原則、結構原則、處理原則、內設原則、敏感度原則與儲存原則。一、SAS LOG視窗提供之訊息：說明、警示與錯誤三種訊息，而錯誤訊息又包括了：查無此變數、語法錯誤、輸入錯誤與執行過程錯誤等。三、資料檢視之方法：PROC MEANS、PROC FREQ、PROCPRINT、PROC UNIVARIATE與PROC PLOT。四、處理大量資料的方法：不需要的變數勿保留、DROP或KEEP指令置於適當位置、SORT指令之使用、CLASS指令代替BY指令、WHERE指令代替IF指令、ELSE IF代替IF、資料之合併與ORTIONS之使用等。五、研究者須對欲分析資料的內涵有所瞭解，務使資料可靠、正確，則後續之分析結果才具可信性。

The main issues of this study were to state principles of statistical data handling under SAS system. The principles may prevent researchers to get the results of ＂garbage in， garbage out＂ in data processing. For main issues were as follows: First， principles of statistical data hand-ling: automicity principle， appropriate data principle， data control principle， data manipulation principle， default principle， data sensibility principle and save output principle. Second， information provided by SAS LOG window including notes， warning and error. Error message such as: variable not found， syntax error， invalid data error and execution-time error. Third， methods of data screening: PROC MEANS， PROC FREQ， PROC PRINT， PROC UNIVARIATE and PROC PLOT. Fourth， methods of working with large data set: drop all unnecessary variables， use a DROP (or KEEP) option on SET statement， do two-level sort， using a CLASS statement instead of a BY statement with PROC MEANS， use the WHERE statement instead of an IF statement， use ELSE IF instead of multiple IF statements， use PROC APPEND to add new data to a large file and use OPTIONS statement properly.