[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Child Maltreatment Rates as an Outcome Indicator
I would argue that entrance cohorts have a significant advantage for
agencies that may be interested in knowing whether the risk of recurrence
declines over time - while intervention is being provided.
Please see the following recent articles where we use survival analysis to
following families prospectively over five years and to predict recurrences
while services are being provided.
DePanfilis, D., & Zuravin, S. J. (1999). Predicting child maltreatment
recurrences during treatment. Child Abuse and Neglect, 23, 729-743.
DePanfilis, D., & Zuravin, S. J. (1999). Epidemiology of child maltreatment
recurrences. Social Services Review, 73, 218-239.
Diane DePanfilis, Ph.D., MSW
Assistant Professor
University of Maryland School of Social Work
525 West Redwood St.
Baltimore, MD 21201
410-706-3609, 410-706-6046 (fax)
ddepanfi@ssw.umaryland.edu
>>> "Greg Tooman" <GREG@americanhumane.org> 03/20 1:45 PM >>>
At the American Humane Association, we've been applying survival analysis
(e.g. life tables) to recurrance data with positive results.
Survival Analysis was developed for biology, and as the name implies, it
measures 'survival' rates past the point of some 'terminal event.' In a
child welfare application, survival analysis can be applied to recurrance in
the following way: the variable of 'number of days from close of service to
recurrence of maltreatment' can be measured, with the recurrence coded as
the 'terminal event.' Therefore, if you have an 80% survival rate for a
population, that means that there was a 20% recurrence rate.
Survival analysis is very handy for any analysis where time is a dependent
variable and you have a wide range of start points. When (within your
measurement period) a start point begins can be a key confound, since cases
that you begin to observe close to the end of your observation period
obviously don't have enough time to 'mature' before you are done measuring.
In the recurrence example, think of a case that closed two months before you
began your data analysis compared to a case that closed nine months before.
The case that closed nine months before your data analysis has had more time
to have a recurrence than the one only given two months. Survival can also
account for multiple factors of interest - like for instance treatment
types or perhaps number of children in the household - that could influence
recurrence.
One key 'sticking point' is whether you are measuring entry cohorts (select
cases for analysis based on when they entered the system) or exit cohorts
(select cases for analysis based on when they exit the system). Entry
cohorts are preferable methodologically, but since you're measuring
recurrence it makes more logical sense to pick exit cohorts, since the
variable of interest - recurrence - can't occur until the case is closed
first.
Measuring recurrance of maltreatment in an unbiased manner can be tricky,
but extremely important in terms of determining programatic success. Hope
this adds something coherant to the discussion. Unfortunately, many of
these analytic problems don't lend themselves easily to verbal description.
Gregory Tooman, M.A.
Research Analyst
Children's Services
American Humane Association
>>> JTabin@aol.com 03/20/00 07:08AM >>>
Sharon,
This message provides the additional explanation I promised on longitudinal
and cross-sectional analysis.
In the meantime, some of Mark Chaffin's comments have touched on similar
issues, but in different ways.
I should begin by clarifying one of my key assumptions. I made my original
comment assuming the purpose of your analysis is evaluation of your
program.
In other words, I assume you are looking at ways of determining whether
your
program is effective and whether it is improving.
Measuring improvement is easier than measuring effectiveness,
unfortunately.
Simple longitudinal comparisons can reflect improvement. Longitudinal
means
'over time' -- this time period compared to the previous time period, etc.
Effectiveness might be judged by comparing your program to similar
programs.
That is what cross sectional analysis means. In business today, we call
something similar benchmarking: studying the process of successful
operations
in order to improve our own process.
Change analysis -- that is, measurement at period x compared to measurement
at periods x-1, x-2, x-3, etc. -- can be very helpful; it washes out lots
of
minor variables and makes the differences in how different variables are
constructed far less important. Once you determine change in two
dissimilar
measurements over the same time periods, it is often possible to look at
change in variable a (delta a) compared to change in b (delta b) quite
usefully.
This may or may not be useful with your particular data. The first thing I
would want to know is whether or not there is a fairly constant
relationship
between the county's maltreatment rate and your total number of mothers in
program. If the county's maltreatment rate various from period to period,
but your program can serve only a fixed number of mothers (as opposed to
varying with changing needs), then there is no basis whatsoever for
comparing
the two sets of data. On the other hand, if your program must respond to
changing needs, such analysis could be very useful. One might test the
hypothesis that, as environmental demands increase (reflected in county
data), program failure increases (reflected in your program data).
Finally, one comment on Mark Chaffin's choice of language; Mark says a
particular comparison "is probably not a fair comparison." In data
analysis,
the word 'fair' introduces a confusing issue. Data analysis should be
concerned with better understanding of reality and better decision making
going forward. Words like 'useful' and 'reasonable' and 'logical' and
'realistic' suggest good analytic thinking to me. As soon as someone cries
'unfair,' I suspect we are dealing with punitive judgments of past
performance, a practice that fogs up clear and effective thinking and
planning.
That is a lot -- but still leaves much open for discussion. If you think I
can help you further, let me know.
Janet Tabin