Use longer trends in Trend, Achievement and Loss?

[pboisver]

Motivation

We can allow for variable length trends. A trend is defined as strictly monotonic increasing performance level including the current month with a minimum of three months of performance levels (two increases in performance).

Start with Trend

A positive Trend is detected when

• there an increase in performance rate from last month to this month
• there an increase in performance rate from last-last month to last month
• These first two criteria imply a minimum 3 months to define a trend.
• There must be data for the current month (>= 10 cases)
• "last month" is defined as the most recent month for which there is valid data prior to the current month
• "last-last month" is defined as the most recent month for which there is valid data prior to the last month

examples (positive trend):

trend signal?	Sep	Aug	Jul	Jun	May	trend length	slope
yes	10	9	8	x	x	2	1
yes	10	9	9	x	x	1	2
no	10	10	9	x	x	-	-
no	10	10	10	x	x	-	-
no	10	11	8	x	x	-	-
no	10	9	11	x	x	-	-
no	10	11	12	x	x	-	-
yes	10	missing	9	8	x	2	.67
yes	10	9	missing	8	x	3	.67
yes	10	missing	9	missing	8	4	.5
no (NYI)	missing	10	9	8	x	-	-

*** Questions

• How should we really handle missing data? (for both detection and trend slope and trend length)