INTRODUCTION TO PIVOT CHARTS
Posts 18 and 19 use pivot charts to look at the interactive relationships between variables, including age distribution, number of sessions, length of treatment, parent configuration, format, and termination type. The search is for interesting patterns that emerge as the data gets organized by these charts.
The first question is whether the noticed pattern is a meaningful departure from the norm, or simply an anomaly. For example, take the average CGAS values of the five different parent configurations in Post 14. That result separated both the single mother group and adoptive families from the other three configurations. The two had similarly low averages. The single mother’s group routinely encounter difficult life circumstances that could account for their disparity in clinical results. That difference was appearing in other study findings as well. Upon further investigation, though, the adoptive family group seemed to be a random collection of particularly unusual cases for that particular clinical population which in turn presumably led to lower outcome averages. Both groups will be discussed further in the Sub-Group Study Section, but one is a departure from the norm as established by other results, and the other is an anomaly.
In reviewing these charts and discovering patterns that separate particular groups from others, consider a tier of possibilities. Does the observed pattern seems potentially meaningful? Does a broader pattern of results involving the identified group in question exist within the data set? Can similar findings in other studies be located? Is the overall data set sufficiently large to run statistical significance tests? The answer here is ‘no’. However, do the findings nevertheless warrant some kind of clinical consideration toward process adjustment(s) to see if the results of therapy can be positively effected? Perhaps with enough data over time or in combination with similar data from other sources, the sample may become of sufficient size to begin formal statistical evaluations.
This investigative research process is called “data dredging”. The term refers “to the use of data mining methods to sample parts of a larger population data set that are too small for reliable statistical inferences to be made about the validity of any patterns discovered.” (Wikipedia entry on Data Mining).
The quote continues: “These methods can, however, be used in creating new hypotheses to test against the larger data populations”. A definitive caution in the Wikipedia entry on “Data Dredging” itself states that “dredging would be a misuse of data analysis”.
As an aside, the thinking behind dredging also represent examples of abductive reasoning. If the underlying pattern is of sufficient concern but cannot be statistically substantiated, experimenting with minor clinical process changes, additions, or deletions within the bounds of ethics codes can still be warranted. Examine the results, re-hypothesize and reformulate as indicated, examine the results, re-hypothesize, etc., until consistent, positive results are occurring.
As the lowest form of statistical truth searching, lower than data mining which at least has a theoretical target to seek, the name ‘dredging’ does give off a connotation of being impure. But, data mining and dredging can still be a part of the search for truth.
These two posts can be seen as a small educational opportunity. If the reader is so inclined, spend a few minutes perusing through the presented charts. They begin with a series of five with the primary variable being Age Distribution, then four by the Number of Sessions variable that begins Post 19, then three via the Length of Treatment, and so on through the post. Pick out what appear to be patterns that do not seem to fit with the overall distribution of the particular chart. I am going to be presenting and discussing my own observations at the end of each primary variable series.
One socio-political factor to consider has to do with how mental health services will be allocated in the future. The main point of dredging herein is to enhance treatment through understanding case characteristics and management, process effectiveness, and developing strategies and tactics pinpointing certain clinical groups. In the broader picture, though, rising national mental health needs combined with increasing proportions of the population unable to afford mental health treatment may lead to broad managed care approaches that limit the numbers of services available per person per problem per period of time (commonly a year). Some of the data herein does tangentially illuminate problems inherent in capping mental health care.
Age X # Sessions
Age X Length of Treatment
Age X Parent Configuration
Age X Format
Age by Termination Type
Age X #Sessions – As indicated before, the length of treatment in terms of both session usage and time in treatment are self-determined, i.e. the primary decision rests with the client and parents. The clinician generally plays an advisory role. With those conditions in mind, the moderate upward slope of the age X # of sessions from the oldest to the youngest suggests that session usage in inversely proportional to age. The youngest tended to have more sessions than the oldest, and the inferred line of significance appears to be straight or nearly straight. If this data were available on a national basis for general outpatient child and family therapy, a similar finding would not be surprising, but is only conjecture here.
X LoT – The significance line here would be close to level. The difference between this finding and the one above, both of which are measuring roughly equivalent data, would likely be the added time of the split process cases bring to the total time. Neither finding is particularly important to this study.
X Parent Configuration – The disproportionate number of 12 – 16 year old boys in the single mother parent configuration could be seen as a poignant reminder of sole parenting’s inherent stresses, given the difficulties many families experience as individuals of this age group emotionally and dramatically burgeon toward adulthood.
X Format – The concentration of 12 – 16 year olds in conjoint and split session processes is also related to the single mother group, where eight of ten were in conjoint and one other was in the split-session format. Also, the 12 – 13 year age group in particular stands out here and in previously noted findings as one which may be particularly well-suited to the conjoint format, having started with lower average CGAS scores and a relatively high gain average that used a low average number of sessions.
Thanks to Alan Leider, MD,
1920 – 2010
Consultant, teacher, mentor