12986274 - Laserfiche WebLink

Home Browse Search

Consequently,most national estimates analyses allocate unknowns. This is done by using scaling ratios defined by NFPA survey estimates of totals divided by only those NFIRS fees for which the dimension in question was known and reported. This approach is equivalent to assuming that the fires with unreported characteristics,if known, would show the same proportions as the fires with known characteristics. For example,it assumes that the fires with unknown ignition factor contain the same relative shares of child-playing fires, incendiary-cause fires, short circuit fires, and so forth, as are found in the fires where ignition factor was reported. Rounding Errors The possibility, of rounding errors exists in all our calculations. One of the notes on each table indicates the extent of rounding for that table,e.g.,deaths rounded to the nearest one, fires rounded to the nearest hundred,property damage rounded to the nearest hundred thousand dollars. In rounding to the nearest one,functional values of 0.5 or more are rounded up and functional values less than 0.5 are rounded down. For example,2.5 would round to 3,and 3.4 would round to 3. In rounding to the nearest one,a stated estimate of 1 could be any number from 0.5 to 1.49,a roughly threefold range. The impact of rounding is greatest when the stated number is small relative to the degree of rounding. As noted,rounding to the nearest one means that stated values of 1 may vary by a factor of three. Similarly,the cumulative impact of rounding error-the potential gap between the estimated total and the sum of the estimated values as rounded-is greatest when there are a large number of values and the total is small relative to the extent of rounding. `, Suppose a table presented 5-year averages of estimated deaths by item first ignited,all rounded to the nearest one. Suppose there were a total of 30 deaths in the 5 years,so the total average would be 30/5 =6. In case 1, suppose 10 of the possible items first ignited each accounted for 3 deaths in 5 years. Then there would be 10 entries of 3/5=0.6,rounded to 1,and the sum would be 10, compared to the true total of 6. In case 2,suppose 15 of the possible items first ignited each accounted for 2 deaths in 5 years. Then there would be 15 entries of 215=0.4,rounded to 0, and the sum would be 0, compared to the true total of 6. Here is another example: Suppose there were an estimate of 7 deaths total in 1992 through 1996. The 5-year average would be 1.4,which would round to 1,the number we would show as the total. Each death would represent a 5-year average of 0.2. If those 7 deaths split as 4 deaths in one category(e.g., smoking)and 3 deaths in a second category(e.g.,heating),then we would show 4 x 0.2 =0.8 deaths per year for smoking and 3 x 0.2=0.6 deaths per year for heating. Both would round to 1,there would be two entries of 1, and the sum would be 2,higher than the actual rounded total. If those 7 deaths split as 1 death in each of 7 categories(quite possible since there are 12 major cause categories),then we would show 0.2 in each category,always rounding to 0, and the sum would be 0,lower than the actual rounded total. The more categories there are,the farther apart the sum and total can—and often do -- get.