Biology Reference
In-Depth Information
All non-compliances
Non-compliances with public health impact
3.0
2.5
2.0
1.5
1.0
0.5
0.0
7 14
Size of Time Window for Evidence Aggregation (days)
28
Figure 9.6
Results of lift analysis executed to characterize relationships between regulatory non-
compliances and the outcomes of microbial testing of food samples for
Salmonella
at food
factories.
These outcomes are accumulated in a 2-by-2 contingency table across mul-
tiple days and multiple factories. Then, a range of high-level statistics about
the relationship between non-compliances and results of
Salmonella
tests,
such as odds ratio, recall, false positive rate, lift, etc., can be derived from that
table. Lift statistic estimates the ratio of conditional probability of observing
a positive
Salmonella
test result in a randomly selected factory, given that
it recently recorded non-compliances, to the prior probability of observ-
ing the
Salmonella
positive without taking non-compliances into consider-
ation. Under null hypothesis of no relationship between non-compliances
and
Salmonella
test results, lift should equal 1.0. Any lift value significantly
greater than that suggests a positive correlation.
Figure 9.6 depicts results of computing lift statistic for two different sub-
sets of non-compliances and three widths of temporal windows of evidence
aggregation and the outcome window of seven days. The lift statistic charac-
terizes here the utility of observing recent regulatory compliance records in
order to predict the risk near future microbial problems. White bars depict
performance attainable using occurrences of all kinds of non-compliances
in making predictions. The gray bars correspond to using just a subset of
non-compliances that are believed by the food safety experts to potentially
bear public health consequences. Being more selective about data aggrega-
tion pays off, as the observed lift values exceed those obtained without dis-
criminating between the different types of recorded non-compliances. Given
that the data of public-health gravity is just a subset of the complete record
