"There is also one particularly needed elaboration of the model (which is not to say that it doesn't cry out for elaboration in many other directions). This is the relation between college filtering and on-the-job filtering. Once an employee has been hired, the employer can gradually draw on more directly obtained information to determine his productivity. However, this filtering may be costly. To the extent that the employer does filter and does so accurately, the Value of the college filter is reduced. The employer pays the average product of a group with given educational achievement only during the period before his own filter has become effective. Conversely, however, an increase in the college population will mean (and has meant) a depreciation in the quality of non-college students (this is not necessarily the same as a decrease in the quality of college students)."