Table 7.

Imputed bootstrap model selection results for the outcome measure persistent disability

	most frequently selected models					rank
Predictors*	1	2	3	4	5	MI-5+B	MI-5
concomitant lower back pain	X	X	X	X	X	1	1
longer duration of complaints	X	X	X	X	-	2	2
both shoulders afflicted	X	X	X	X	X	3	3
inability to perform daily activities	X	X	X	-	X	4	4
higher scores for somatisation	X	X	X	X	X	5	5
higher scores for external locus of control	X	X	X	X	X	6	6
more disability at baseline	X	X	X	X	X	7	7
older age	X	X	-	X	X	8	8
shoulder complaints in the past year	-	X	X	-	X	-	-
concomitant lower extremity pain	-	-	-	-	-	-	-
Count	91	77	56	54	52
%	3.6	3.1	2.2	2.2	2.1

*             - only those predictors that appeared in ≥40% of the first bootstrap model selection step are presented

rank        - the order of appearance of predictors in the derived models arranged by their predictive ability (regression coefficient estimates)

MI-5+B    - the multiple imputation based bootstrap selected model (i.e. the most frequently occurring combination of predictors in 2500 replicate data sets of the second bootstrap model selection step)

MI-5        - the multiple imputation based model using 5 imputed data sets was also the most frequently occurring combination of predictors in the 2500 bootstrap replicate data sets

Count      - the number of times the model was selected in the 2500 replicate data sets of the second bootstrap model selection step.