Table 3.
How often are predictions of blood use better for random than for purposive sampling strategies?
| RAND versus: | MAXVAR | LARG | ||||
|---|---|---|---|---|---|---|
| RBC | FFP | PLT | RBC | FFP | PLT | |
| Lower hospital level prediction error for RAND | 0 % | 3 % | 14 % | 100 % | 35 % | 84 % |
| Lower national level prediction error for RAND | 25 % | 3 % | 79 % | 100 % | 81 % | 87 % |
| REGVAR versus: | MAXVAR | LARG | ||||
| Lower hospital level prediction error for REGVAR | 0 % | 6 % | 25 % | 100 % | 44 % | 90 % |
| Lower national level prediction error for REGVAR | 20 % | 4 % | 79 % | 100 % | 79 % | 88 % |
| 2REG versus: | MAXVAR | LARG | ||||
| Lower hospital level prediction error for 2REG | 0 % | 0 % | 0 % | 100 % | 9 % | 52 % |
| Lower national level prediction error for 2REG | 24 % | 3 % | 83 % | 100 % | 70 % | 88 % |
Percentage of all simulations that the random strategies (RAND, REGVAR and 2REG) outperform the purposive strategies (MAXVAR and LARG) in terms of hospital and national level prediction error for n = 12 (n = 4 per hospital type)