In
this case we made use of a paired t-test to establish that there was
a statistically significant change in the mean number of veterans
waiting 31-60 days for an appointment from a sample of Southeastern
US veterans medical centers. A paired comparison was the appropriate
analysis since there is a natural pairing between each hospital’s
2015 and 2016 backlogs and the VMCs are not homogeneous.
Stakeholders are hoping
for improvement in the performance of VMCs with respect to delays
in treating veterans in need of care. However, it is important to
formulate the problem statement and hypothesis test in a way that
will detect a change in performance, either improvement or degradation,
over the course of a year. While measures were taken to reduce the
backlogs, they may not have been effective. Other possible reasons
for the increased backlog could be increased demand for service or
decreased staffing levels (a problem that has also be reported in
the VA medical system).
The test of hypothesis
established the statistically significant increase, but examining
the graphs showed that Fayetteville, NC was the only VMC that achieved
a substnatial reduction in backlog. Outliers can influence the outcome
of a hypothesis test. A statistically significant difference can
be obtained with the outlier included in the analysis, but excluding
the outlier results in a difference that is not significantly different
and vice versa. In small sample sizes, outliers can be particularly
influential. From a statistical viewpoint, outliers should be investigated
to look for possible causes such as a data collection error. From
the perspective of the problem domain, investigating an outlier that
is not found to be in error may provide valuable insights. For example,
Fayetteville may be doing something differently than the other VMCs
that is effective in reducing the backlogs.
In this case we analyzed
only one facet of the problem, the change in the 31-60 backlog. The
VA provides information on the number of veterans waiting both shorter
and longer periods of time for appointments. While there appears
to be an increase across the Southwest US, the data set contains additional
variables that can be analyzed. For example, are there differences
between VISNs or states? How does bed capacity affect the backlog?
More detailed analysis should be conducted so that a complete understanding
of the problem is obtained. To improve performance it is important
to understand if problems are systematic or related to another factor
such as VISN, state, or bed capacity. This will enable targeted improvement
actions to be developed.