departures from a firm are hidden under another category, such as ''fired" or "quit," when in fact the departure was due to AIDS. If people systematically return to their families in rural areas to receive care, this scenario is quite likely. In the countries in this sample, attrition due to illness and death ranged from 3 percent of all attrition in Ghana to 13 percent of all attrition in Zimbabwe. Any attrition is costly to a firm, but it seems unlikely that an increment of even 15 percentage points in the attrition rate would be enough to impede seriously a firm's growth. Further examination of this question would require combining the RPED data on worker attrition with data on firm profitability in order to study the impact of the former on the latter.
Unfortunately, the RPED data do not categorize all hires and departures by grade level. However, on the questionnaires for Zimbabwe, Kenya, and Ghana, an additional page of questions was added to gather more complete data on the first nine deaths in each firm. While compliance was not always perfect, of the 592 workers who had left because of death, some detailed information is available on 258 (44 percent). 6
Table 6-4 presents information on the difficulty of replacing workers by four skill categories. Among the 229 workers for whom data on skill category and the result of the replacement process were available, only 8 were professionals, while the other 221 were grouped into the descending categories of "skilled," "operator," and "unskilled." Focusing on these last three groups, we note that the proportion of deceased workers for whom a replacement had been found as of the time of the interview decreased with increasing skill level, from only 59 percent for skilled workers and 56 percent for operators to 72 percent for unskilled. Conversely, the proportion of openings that the employer was still trying to fill declined from 12.1 percent for skilled workers to 2.4 percent for the unskilled. We must also note that the response "do not plan to replace" was more frequent than the response "still looking" for every skill category, sometimes by an order of magnitude. Whether these responses represent downsizing, a change of job description, an extended and expensive search, or a misunderstanding of the questionnaire is impossible to interpret. Yet the data tend to indicate support for the hypothesis that the labor market is tighter at the top of the skill spectrum than at the bottom.
The last column of Table 6-4 gives, for those positions for which a replacement was found, the number of weeks the employer reported searching until finding the replacement. Here again, though the sample size is small, there is evidence that more search was required at higher than at lower skill levels. There are a number of reasons to expect firms to spend longer searching for higher-skilled
Ghana provided information on 95 percent of its deaths, Kenya on 86 percent, and Zimbabwe on 35 percent. Zimbabwe's lower completion rate was due partly to the fact that some large firms had experienced more than nine deaths. Since the questionnaire had space for only nine questions, the form was not physically able to handle the other deaths.