Dear community,
I want to create a box plot with the data for two treatment groups next to each other. I found a code and adjusted the variable names for my data (and triple-checked them). However, I always get the following error message:
However, the variable "Training_type" definitely exists in the dataset I use. Do you have any other ideas that could explain this behavior?
This is the code I use which throws the error:
DATASET ACTIVATE DataSet1.
USE ALL.
EXECUTE.
/*TG on Error Count Analysis (Hypothesis H1)
DATASET ACTIVATE DataSet1.
/*Boxplot for Outliers
* Chart Builder.
GGRAPH
/GRAPHDATASET NAME="graphdataset" VARIABLES=Training_type Initial_training_error_count_Number_mistakes MISSING=LISTWISE REPORTMISSING=NO
/GRAPHSPEC SOURCE=INLINE.
BEGIN GPL
SOURCE: s=userSource(id("graphdataset"))
DATA: TGnew=col(source(s), name("Training_type"), unit.category())
DATA: mistakes=col(source(s), name("Initial_training_error_count_Number_mistakes"))
DATA: id=col(source(s), name("$CASENUM"), unit.category())
GUIDE: axis(dim(1), label("Training_type"))
GUIDE: axis(dim(2), label("Initial_training_error_count_Number_mistakes"))
GUIDE: text.title(label("Simple Boxplot of mistakes by Treatment group"))
SCALE: linear(dim(2), include(0))
ELEMENT: schema(position(bin.quantile.letter(Training_type*Initial_training_error_count_Number_mistakes)), label(id))
END GPL.
Thank you in advance for your help!
------------------------------
Alexander Farr
------------------------------
#SPSSStatistics