Table 4.
Effects of imputing missing data
| Error | 5 % imputed | 10 % imputed | 25 % imputed |
|---|---|---|---|
| N (%) | N (%) | N (%) | |
| −4 | 0 (0.0) | 0 (0.0) | 1 (0.0) |
| −3 | 0 (0.0) | 0 (0.0) | 17 (0.2) |
| −2 | 6 (0.1) | 27 (0.4) | 126 (1.8) |
| −1 | 225 (3.2) | 437 (6.3) | 893 (12.8) |
| 0 | 6423 (91.9) | 5920 (84.7) | 4691 (67.1) |
| 1 | 316 (4.5) | 560 (8.0) | 1042 (14.9) |
| 2 | 16 (0.2) | 36 (0.5) | 178 (2.5) |
| 3 | 0 (0.0) | 6 (0.1) | 32 (0.5) |
| 4 | 0 (0.0) | 0 (0.0) | 6 (0.1) |
| Total | 6986 (100.0) | 6986 (100.0) | 6986 (100.0) |
| Number of respondents with ≥1 imputed response | 2869 (41.1) | 4621 (66.1) | 6583 (94.2) |
| Number of respondents with error within ±1 | 6964 (99.7) | 6917 (99.0) | 6626 (94.8) |
Errors in Mokken scale scores after removing a percentage of the responses for each asset at random and then re-imputing the data
Error was defined as the difference between the score using imputed data and the score using the original, observed data. The number and percentage of respondents with different magnitudes and directions of error are shown