WORKING GROUP 1 
BIRNBAUM 
63 
Table 5. Percent completeness by photo mode for low photo quality 
Completeness 
Positive 
Transparency 
Negative 
Transparency 
Stereo 
Print 
Non-Stereo 
Print 
Initial 
8 
1 
1 
1 
Final * 21 
* Mode differences significant at P < .05. 
16 
24 
18 
attributable to mode were not found to be significant. Table 4 presents the 
accuracy results for low photo quality. Once again mode differences were not 
found to be significant. However, when we come to completeness at low 
photo quality (table 5) we find a difference attributable to mode that is 
significant at the five percent level of confidence, and this for completeness 
after thirty minutes of work. In view of the fact that eight different analyses 
were conducted, four for low and four for high photo quality, it is not too sur 
prising, on a probability basis, to find one that exhibits differences that appear 
to be significant for the case where in fact no differences exist. The large number 
of analyses therefore casts some doubt on the significance of the finding of mode 
differences for completeness at low photo quality after thirty minutes of work. 
When we examine the effect of photo quality on performance we can un 
qualifiedly say that we have succeeded in reducing photo quality. Without 
exception performance on high quality photos was better than on low quality 
ones as can be seen in table 6. 
* 
Table 6. Percent accuracy and completeness by photo quality 
High Quality 
Low Quality 
Initial Accuracy 
52 
43 
Final Accuracy 
41 
35 
Initial Completeness 
11 
7 
Final Completeness 
30 
20 
Quality differences are significant at 
P < .01. 
It can also be seen that accuracy drops as a function of time. For six-minute 
performance accuracy is higher than for thirty-minute performance. Con 
versely, completeness goes up as a function of time. 
Content differences were significant throughout, i.e., performance varied as 
a function of content. However, this is not too surprising in view of the fact 
that the photograph sets differ from each other in scale, quality, density and 
kind of objects. 
Interpreter confidence 
As a part of their regular work routine, interpreters assign probability indices 
expressing their confidences in each of the identifications that they make. In a 
prior study, a significant difference was found in the mean confidence for 
right and wrong identifications and in favor of the right identifications. The