SECTIONAL TRANSACTIONS.—L. 567 
Mr. F. BarracLoucH.—The reliability of entrance examinations to 
secondary schools (10.30). 
In recent years local education authorities have paid much attention to 
the syllabuses for scholarship entrance examinations, but in many cases a 
corresponding lack of attention to the machinery of the examination has 
vitiated their efforts. 
School records and the position of children in non-provided schools are 
matters requiring further research. 
In making a comparison between the results of scholarship entrance 
examinations and school certificate examinations the methods of analysis 
of variance enable the investigator to test the significance of his results. 
The work of Prof. Valentine regarding the predictive value of scholar- 
ship entrance examinations revealed the need for additional research, and the 
results of an investigation conducted in ‘ non-Valentine’ areas are given. 
The predictive value of intelligence tests is also discussed. 
Mr. F. SaNDON.—Difficulties in using entrance examinations, intelligence 
tests and school results for comparative purposes (10.45). 
Prediction is a question in probabilities, and our estimate of a, the 
ability to profit by education at a secondary school, depends on correct 
statistical procedure. No test, by reason of various kinds of variability, 
will be perfect, and any test will only correlate to some degree with the 
criterion. Even if we have a theoretical correlation table with r, between 
mark and criterion, as 0-85, in the top one-fifteenth two-fifths of the selec- 
tion should not be there and the correlation for selected candidates would 
be computed at 0:3. This illustrates a principle of Karl Pearson that 
examination statisticians have apparently overlooked. Selection modifies 
means, standard deviations and correlation coefficients. In an observed 
case a correlation of 0-67 on feebly selected material was by more and more 
stringent selection reduced through values of 0:07 and —o-16 to —o-93. 
The farther we get away from the measure of the selection the less will be 
the effect on the correlation of correlated measures—the other test will 
always tend to be the better. A correlation coefficient between two measures 
should not be given in vacuo ; the material studied should be revealed and 
special study made of possible selection effects. 
Other statistical and allied difficulties are referred to and some suggestions 
submitted. 
Prof. J. Drever—The comparative reliability of examinations and 
tests respectively (11.0). 
Examination marks are unreliable from two points of view. In the first 
place one and the same examination paper, it is well known, may be 
differently assessed by different examiners, and even by the same examiner 
at different times. This defect might to some extent be removed by objec- 
tive methods of assessment. In the second place school examinations at 
best have been shown to have an extremely low correlation with University 
success even in the same subject. The causes of the unreliability in this 
case would appear to be very complex. The first type of unreliability ought 
not to be present to any considerable extent in the case of intelligence tests, 
and the second is certainly not present in the same degree. 
