Dear All,
Many appologies if you have already read this but Matt has just pointed out
to me that I sent the last attempt in HTML format (I diddn't mean to,
grovelling apologies).
I am preparing to carry out the sampling for my PhD research into how Art
and Music Therapists write about their client outcomes and I am obviously
concerned that from a statistical point of view this sample is valid. Since
my supervisors are Arts people and think that what I propose is fine. I
(being cautious) feel that I would like the opinion of someone who really
knows about sampling to check it out for me. Do any of you have five minutes
to read this and make critical comments?
Many thanks,
Helen Quenet
[log in to unmask]
The Sample Frame and Sampling Method
The sample frame for the study is all the Art and Music Therapy records held
in the archives of the Horizon NHS Trust. From the initial stage of the
study 1283 archived client files were identified. The files originated from
three different hospital sites for the Learning Disabled, covered a thirty
year period from 1970 and were comprised of 825 Art Therapy and 458 Music
Therapy files.
Records were coded according to material they contained; the time period
covered; hospital they came from; how many formal reports they contained and
whether they contained process notes. Following this, the records, which
contained inadequate material for any meaningful content analysis, were
identified:
Type 1-17 no notes in file (file empty).
Type 2- 11 contained 1 non-relevant letter
Type 3- 63 referral form only
Type 4- 127 no notes relating to music or art therapy
Type 5- 179 incomplete assessment
Type 6- 160 assessment only, no further work
Type 7 - 465 very incomplete/sparse notes, no sequence of sessions
Total 1022 with a poor quality of content. Types 1-4 were excluded from the
sample frame since they contained no information relating to music or art
therapy. Types 5-7 were included because they did contain some information
this left 804 files of poor quality for the sample.
This left 261 files (of which 147 were Art and 114 Music) which had good
documentary support for therapy carried out. The difference in the quality
of content in the Music and Art files is probably accounted for by the music
therapists already having cleared uninformative files from their filing
system.
With this information files could now divided up into four groups:
Music Therapy files with good content 114/1065 (11% of total)
Music Therapy files with poor content 269 /1065 (25%)
Art Therapy files with good content 147/1065 (14%)
Art Therapy files with poor content 535/1065 (50%)
When considering strategies for sampling the way in which the desired aims
will be met by any strategy are key. This study has two main aims:
1) To discover if there are indicative criteria for outcomes in Art and
Music Therapy contained within archived records. (Quantitative)
2) To examine how music and art therapists write about outcomes
(Quantitative and Qualitative)
Clearly for the second aim all that was required was that adequate archived
material be accessed to give a clear impression for each of the two
therapies. For the first aim an appropriate sampling strategy was required
that would enable each of the four groups to be adequately represented. A
quota sample taking 70 files at random from each of the four groups yields
different probabilities of any one file being selected:
0.61 for good content music therapy files
0.26 for poor content music therapy files
0.47 for good content art therapy files
0.13 for poor content art therapy files
This means that the sample is weighted towards the good quality files and
any interpretation of results needs to take this into account.
With a sample size of 70 for a confidence level of 95% the confidence
interval for each of the groups is:
+/- 7.3% for good content music therapy files
+/- 10.1% for poor content music therapy files
+/- 8.5% for good content art therapy files
+/- 10.9% for poor content art therapy files
Graph 1 gives the relationship between sample size and confidence interval.
Given the nature of the material a confidence level of +/- 10% would seem to
be reasonable, a sample size of 70 would seem to be a good compromise
between work required and accuracy likely.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|