The above discussion is admittably abstract in that it is almost impossible to quantify precisely most of the costs needed to make the above comparison. However, the regulatory agency should be aware of the fact that the marketplace is also a deterrent with respect to food adulteration. Moreover if the agency uses banning as its method of enforcing standards, consumers lose the opportunity for choice. Consequently consumer protection agencies should consider the costs associated with lost choice as well as the costs of policing, and compare them to the benefits to society associated with consuming more healthful foods and reduced decision making.

Even though one can't measure these costs precisely it is possible to get some handle on the magnitude of the benefit to society and the opportunity cost by looking at the impact of the FDA's actions on the firm's behavior. If in fact the standards maintained by the FDA have little or no influence on the firm's behavior (if there were no agency), then there is no reduction in social costs and probably no restriction in consumer choice. If, however, firms would have much different quality levels without regulatory action, we would expect substantial opportunity costs and the possibility for substantial social gains. Implications of model

What does all this mean? First of all the model points out the trade-offs between attacking the problem of food regulation by increasing consumer information versus setting manufacturing standards and banning. In other words the average food quality will be improved (a) if the consumer is given more quality information, assuming this information causes him to place more weight on quality during the purchase decision process and/or (b) if the FDA sets food standards more stringent than those set by the free market. Thus, it becomes a question as to which of these two methods is most effective. Clearly the answer to this question is at best complex. However, later in the paper we will study the effects of inspection and consumer information in an attempt to provide at least a partial answer of when each approach is most effective.

A second implication of the model is that even with FDA inspections we would expect a range of quality levels and prices within a particular product class and that these quality levels should not be influenced by any one classification of the FDA. In other words, the firm, in response to consumer demand (and possibly FDA standards) positions its product with respect to quality and price and maintains this position unless it senses a shift in either consumer demand or the FDA's standards. Thus, the outcome of a particular FDA inspection should not have any specific influence on the behavior of the firm unless the firm perceives a change on the part of the FDA. This is due to the fact that the firm's quality level decision was made with the expectation that it would be classified periodically by the regulatory agency. Consequently it has already taken into account the actions of the FDA in setting its quality level. No further adjustment is necessary.

The above may seem counterintuitive since one of the main reasons given for having the FDA inspect plants is so that the agency can influence firms to increase the quality level of foods. In a way the plant inspections accomplish this goal in that the inspection frequency and outcome are used by the firms as indications of how the FDA is de

fining and enforcing standards on each quality dimension. However, a firm also "marches to the other drummer," i.e., the marketplace. Thus, it may be advantageous for a firm to produce substandard food (as defined by the FDA) as long as consumer demand for the food is sufficient. This will be true even though the FDA continually classifies the firm to be out of compliance and imposes legal sanctions against the firm (as long as these sanctions don't drive the firm out of business). It is our opinion that under normal circumstances one inspection of a firm will neither change the firm's perceptions of the FDA's standards or drive the firm out of business. Thus the influence of the FDA's visit and classification of the firm will have little or no influence on the behavior of the firm.14

The above discussion, although based on a plausible model of firm and market behavior, is admittably an abstraction of the regulatory process. We now turn our attention to the analysis of the outcomes of a series of FDA inspections in order to better measure the impact of these inspections on the behavior of the firm.

EVIDENCE ON THE EFFECTS OF FDA

Analysis of inspection histories

The FDA maintains detailed computerized inspection histories of each firm inspected. This history contains the inspection dates, the classification of the firm after inspection and actions taken as a result of the inspection (for example, reinspect, send a warning letter, start legal action, etc.) as well as characteristics of the firm such as size, location, etc.

We used this information to find out if firms tended to be classified differently from one inspection to another, and if so, if this change depended on how often a firm was inspected. There are at least two competing hypotheses concerning the behavior of the firm. The first, which is based on the previously mentioned model, predicts that the firm's behavior is not influenced by an inspection or classification, since the firm selects a quality level which balances off the expected costs of regulatory action with the profits associated with consumer demand. The second assumes that a firm will attempt to keep its quality level as low as possible and will only increase its levels as the result of the inspection. However, over time, the new quality level will decay until the next FDA inspection. In other words the act of visiting a firm has a great influence on the firm's behavior. We will concentrate our analysis on attempting to gain some insights into the validity of these two competing hypotheses.

The records for each firm were examined for a four year period between 1968 and 1972 and when possible the following information was compiled: (1) The firm's status at the time of last inspection; (2) the time between inspections; and (3) the firm's status after the present inspection.

14 Recently the FDA has been prosecuting the officers of the firm for criminal violations. For example in 1976, 42 firms received notices to officers that the agency is contemplating taking criminal action, 26 cases were forwarded to the U.S. attorney and 20 cases were filed in U.S. District Courts. This action occurs only after a firm has received two warnings letters. By personalizing the cost of noncompliance we suspect the FDA's classification may alter the management's expectations of the costs and thus the firms' quality levels. Theoretically these costs should have already been taken into account. However, we suspect many executive officers are not aware of the practices at the plant level. Consequently the FDA's decision to go after upper management for criminal violations increases the costs to the firm's decision maker after the inspection. In these cases we might expect substantial changes after the first inspection.

We used this information to compile a "transition matrix" for plants in two different food industries classified by time between inspections. The transition matrix shows the proportion of firms moving from one classification to any other classification. The results are displayed in Tables 1 and 2.

For example, the upper left hand entry in Table 1 is .64. This figure represents the proportion of plants initially found to be in compliance (IC) who were re-inspected within six months and who were reclassified IC on the second inspection. Likewise .22 represents the proportion of plant initially classified IC who were re-inspected within six months and who were then classified voluntary action indicated (VAI). Carrying the example through, the .14 represents those plants initially classified IC who were re-classified after the second inspection (which took place less than seven months after the first) as official action indicated (OAI).

TABLE 1.-PROBABILITY OF CLASSIFICATION AS A FUNCTION OF LAST CLASSIFICATION AND TIME BETWEEN INSPECTIONS FOR INDUSTRY A

TABLE 2.-PROBABILITY OF CLASSIFICATION AS A FUNCTION OF LAST CLASSIFICATION AND TIME BETWEEN INSPECTIONS FOR INDUSTRY B

Industry A is representative with respect to the FDA policies toward most food industries in that FDA has not altered its industry standard or procedures during the period of analysis. Industry B. however, is atypical since the regulatory agency during the last two years of analysis instituted a special intensive random inspection program within the industry, and in general let it be known that they were cracking down. Thus, within Industry B, the FDA increased a firm's probability of being inspected and possibly the

probability of being classified out of compliance. In terms of the model presented above, the FDA increased the firm's expected costs of being found out of compliance. This change in expected costs would imply that some firms might shift their contamination level to reduce their expected inspection costs.

The first thing which is evident from Tables 1 and 2 is that the conditional transition probabilities for both industries do not seem to be related in any systematic way to the time between inspections. In other words, there does not seem to be a trend for plants being found OAI (or VAI) as the time between inspections is increased after controlling for the last classification. This lack of a trend is counter to the hypothesis that firms gradually fall out of compliance (i.e., gradually produce poorer quality foods). However it is consistent with the hypothesis that a firm sets a particular contamination level based upon the market (and possibly the expected fixed costs imposed by the FDA) and keeps this level. Under this assumption of a fixed contamination level we would attribute the fact that firms are classified differently from the previous classification to the measurement error associated with the classification procedure.

The above analysis only considered two factors, namely the time between inspections and the status as determined by the previous inspection. There is at least one other explanation for the lack of a relationship between the transition probabilities and time between inspections. Suppose that firms really do clean up as a result of inspection and then gradually drift out of control until inspected again by the FDA. Also assume that the FDA adjusts its sampling plan so that on the average at the time of inspection the probability of being classified in any new state given a particular old state is fixed. This logic would produce transition probabilities which did not vary with length of time between inspections.

Fortunately, it is possible to test the validity of this hypothesis since these assumptions also imply that the partial effect of time between inspections (i.e., after adjusting for all other factors used by the inspector to determine when to inspect) on the compliance rate would be negative. Two different analyses were performed to test out this downward drifting quality level hypothesis for Industries A and B.15 Both analyses indicated that there was no significant influence of time between inspections on classification. Thus, even after controlling for factors such as previous classification, size of firm, action taken last time, etc., there seems to be no increasing tendency for a firm to be classified out of compliance as the time between inspections increases, again invalidating the decaying quality hypothesis.

Another interesting result of these analyses was that the proportion of firms classified IC, VAI and OAI within an industry tended to differ from district to district even after adjusting for size of firm, past history, etc.16 This implies that two firms with identical characteristics and inspection histories but in different districts have different probabilities of being classified in a specific category. There are many explanations for these observed differences. First, it should be remembered that the classification procedure is a complex task.

15 The analyses used were multiple regression and linear discriminate analysis.

18 This adjustment was done through multiple regression analysis Specific results can be found in Hinich and Staelin (1976).

83-944-78-29

Thus, different districts may use different rules or procedures to arrive at the classification. Also, district offices may use the same procedures but have different cut-off points for classification into one of the three groups. In other words they all would agree that a given write-up indicated a particular level of adulteration or misbranding, but they would not agree as to the proper classification for the given level. This disagreement may in part reflect the fact that districts rely on different circuit courts for enforcement, and these courts at times have different views with respect to food adulteration. No matter what the reason, there seem to be differences between districts.

In summary, we found that the time between inspections had almost no influence on the classification of the firm after controlling for previous classification, and that even after controlling of other factors there were strong district effects with respect to the average classification of a firm within an industry.

Recommended changes in FDA procedures

What sorts of conclusions can one draw from the preceding discussions? First there is some evidence that any one FDA inspection has little influence on the operation of the firm with respect to food adulteration or misbranding. Instead, firms select a particular quality level which is maintained over long periods of time. That is not to say that if the FDA were to stop inspections that firms would not alter their food standards, since discontinuation of inspection would alter firms' perception of the expected costs associated with being found out of compliance. However, it is our opinion that small changes in the field force would not greatly affect the firm's perceptions of expected costs and consequently would have little effect on the average food contamination level within the United States.17

A second conclusion is related to how the FDA implements the law, i.e., how it inspects and classifies firms. We have mentioned more than once that the classification procedure has much room for human judgment and thus human error. Even though the standards are reasonably precise, the classification procedure is not. We suggest that the FDA spend more effort in determining a scoring scheme which could be used for classification. Thus, inspectors would be required to score firms on specific attributes of the manufacturing process. For example the inspector might rate the firm on the condition of the raw materials storage facility (by visually inspecting for insects and rodents), the sanitation practices of the employees and the quality control facilities of the firm. These factors could then be used to determine a statistical relationship between the score on a particular attribute and the adulteration level of the final product. That is, the scores would provide a profile of the firm's operation which could then be related to the status of the output (i.e., the adulteration level or safety level of the food). Based on this evidence FDA could channel its efforts on those manufacturing procedures which were related to adulteration of the final product which pose health hazards to consumers.18 The result would

17 In fact the most important aspect of the problem is not to increase (or decrease) FDA's activities, but to increase (or decrease) the firm's perceptions of the costs associated with the FDA's activities. Our estimate of small changes is plus or minus 25 percent.

18 It is also possible to consider final product adulteration which does not pose a health hazard to the consumer. It is our view that the FDA should not pay much attention to this type of adulteration. However, our suggestions with respect to finding a relationship between input measures and final output hold, no matter what criterion is used for adulteration.