Hutan
Senior Member
- Messages
- 1,099
- Location
- New Zealand
There's something a bit weird with the imputation of missing data. If you look at the supplementary material on imputation, it appears that,
for specialist services data at both 13 and 16, only around 15% of the 1680 data points had BMI information;
for ALSPAC data at both 13 and 16, only around 40% of the data points had CFS data and only about 40% had BMI information.
The supplementary material says missing data was imputed using auxiliary variables due to, among other things, 'their strong hypothesised association with CFS/ME'. Auxiliary variables used included
Family Adversity Index
Life Difficulties
School Absences in Year 11
Academic attainment
Anxiety and depression measures
Different variables were used for the imputation in each data source (Alspac and Specialist).
(There was a mention of 99 data points created by imputation for each of data sources, but I'm not sure how that works as they would have had to create a lot more data points than 99 to make up for the missing data.)
But anyway, the supplementary material gives the results from the 'complete case analysis' which presumably is the analysis of the data points that had no missing data. Here's the complete case results (with the results that were reported in the main body of the paper in brackets for comparison).
...............................Alspac No CFS ........... Alspac CFS...............Specialist CFS
13 years
Mean BMI..................20....................................20..............................21...............
% obese....................3.57 (4.18).......................3.92 (3.72).................9.23 (9.28)
16 years
Mean BMI.......... .......21.....................................21..............................22.............
% obese...................3.94 (4.46).........................6.00 (5.46).................6.8 (16.43)
So, the differences between the three groups look a lot less impressive in the un-imputed data. Instead of an upward trend from 9% to 16% obese over three years in the specialist group, there is a downward trend from 9% to 7%.
That makes a big difference - there isn't evidence in the complete data points that the CFS children are growing more massive as each year passes.
The imputation process is not well explained. The raw data however is reported as being available - so that is something.
My conclusion: there may possibly be a very small difference between the BMI's and obesity prevalence of non-CFS and CFS adolescents but there's enough murkiness about what was actually done with the data that I would not be getting excited until someone without an agenda analysed the data afresh.
ETA: There seems to be a few very odd things about this paper, so I may well have misunderstood something. But am too tired to be bothered to look again.
for specialist services data at both 13 and 16, only around 15% of the 1680 data points had BMI information;
for ALSPAC data at both 13 and 16, only around 40% of the data points had CFS data and only about 40% had BMI information.
The supplementary material says missing data was imputed using auxiliary variables due to, among other things, 'their strong hypothesised association with CFS/ME'. Auxiliary variables used included
Family Adversity Index
Life Difficulties
School Absences in Year 11
Academic attainment
Anxiety and depression measures
Different variables were used for the imputation in each data source (Alspac and Specialist).
(There was a mention of 99 data points created by imputation for each of data sources, but I'm not sure how that works as they would have had to create a lot more data points than 99 to make up for the missing data.)
But anyway, the supplementary material gives the results from the 'complete case analysis' which presumably is the analysis of the data points that had no missing data. Here's the complete case results (with the results that were reported in the main body of the paper in brackets for comparison).
...............................Alspac No CFS ........... Alspac CFS...............Specialist CFS
13 years
Mean BMI..................20....................................20..............................21...............
% obese....................3.57 (4.18).......................3.92 (3.72).................9.23 (9.28)
16 years
Mean BMI.......... .......21.....................................21..............................22.............
% obese...................3.94 (4.46).........................6.00 (5.46).................6.8 (16.43)
So, the differences between the three groups look a lot less impressive in the un-imputed data. Instead of an upward trend from 9% to 16% obese over three years in the specialist group, there is a downward trend from 9% to 7%.
That makes a big difference - there isn't evidence in the complete data points that the CFS children are growing more massive as each year passes.
The imputation process is not well explained. The raw data however is reported as being available - so that is something.
My conclusion: there may possibly be a very small difference between the BMI's and obesity prevalence of non-CFS and CFS adolescents but there's enough murkiness about what was actually done with the data that I would not be getting excited until someone without an agenda analysed the data afresh.
ETA: There seems to be a few very odd things about this paper, so I may well have misunderstood something. But am too tired to be bothered to look again.
Last edited: