JOURNAL OF POLITICAL ECONOMY This assumes, as mentioned earlier, that each firm expects the wage rate paid by other firms to be constant at least for the duration of the wage negotiated. The rationale of(10), stated loosely, is that the average wage rate will rise(fall) if all firms want to pay a wage higher(lower)than other firms. B It is assumed here that firms in the aggregate adjust their wage only gradually in the direction of the average desired differential; other- wise u and u would be implied to adjust instantaneously to make a' continuously. The gradualness might come from the administrative and psychic cost of changing wage rates that causes wage rates to be changed only intermittently or periodically; if these wage negotiations are stag gered across firms or across workers, then the average wage will move more or less smoothly as indicated In addition, perhaps uncertainty of the firm that the "desired "wage differential, if instituted, would have the desired effect upon turnover costs will induce a cautious, gradual response in the individual firms wage decision As for the postulated shape of the m function, the signs of the derivatives course fundamental to the theory. The excess-demand theory, which is a special case assumes that the second derivatives are zero with m2 =-my= constant>0. My weaker restrictions on the second derivatives in(9b )are inessential they affect only the curvature of the augmented Phillips curve which I shall derive The inequality mu20 meaning that a* decreases with the unemployment rate at a non-increasing rate, vacancy rate constant, is plausible if, as the data suggest (Eagly, 1965), the quit rate is likewise convex with respect to the unemployment rate. The inequality m2220 assumes"rising marginal costs "to the firm of filling vacancies by means other than raising its wage differential. Finally m12 s0 makes sense if it takes a larger increase of the firms wage differen- tial to facilitate the filling of some fraction of a given increment in its vacancies the smaller is the unemployment pool from which workers can conven- iently be drawn. The curve labeled m(u, u)=0 in Figure l gives the com binations of u and u that make 4*=0. Its slope, being -m2/ ml, is necessarily positive, but the size of that slope and the curvature are in determinate and of no qualitative consequence, To the right of this locus and to the left△*<0 In the United States and most other countries, satisfactory vacancy data are still unavailable. I shall couple the above model with a theory of labor turnover or employment dynamics, along lines suggested by Lipsey(1960),in order to derive testableimplicationsofrelations amongeasily observable data The absolute time rate of increase of the aggregate number of persons employed, denoted N= dN/dt, consists of the number of persons hired 18 Stability of the aver exist firms content with /hat counts for the average wage mo is the weighted weighted average differential, say A, which necessarily equals zero)688 JOURNAL OF POLITICAL ECONOMY This assumes, as mentioned earlier, that each firm expects the wage rate paid by other firms to be constant at least for the duration of the wage negotiated. The rationale of (lo), stated loosely, is that the average wage rate will rise (fall) if all firms want to pay a wage higher (lower) than other firms.18 It is assumed here that firms in the aggregate adjust their wage only gradually in the direction of the average desired differential; other￾wise z. and u would be implied to adjust instantaneously to make A* = 0 continuously. The gradualness might come from the administrative and psychic cost of changing wage rates that causes wage rates to be changed only intermittently or periodically; if these wage negotiations are stag￾gered across firms or across workers, then the average wage will move more or less smoothly as indicated. In addition, perhaps uncertainty of the firm that the "desired" wage differential, if instituted, would have the desired effect upon turnover costs will induce a cautious, gradual response in the individual firm's wage decision. As for the postulated shape of the m function, the signs of the derivatives in (9a) are of course fundamental to the theory. The excess-demand theory, which is a special case, assumes that the second derivatives are zero with m2 = -ml = constant > 0. My weaker restrictions on the second derivatives in (9b) are inessential; they affect only the curvature of the augmented Phillips curve which I shall derive. The inequality m,, 2 0, meaning that A* decreases with the unemployment rate at a non-increasing rate, vacancy rate constant, is plausible if, as the data suggest (Eagly, 1965), the quit rate is likewise convex with respect to the unemployment rate. The inequality m2, L 0 assumes "rising marginal costs" to the firm of filling vacancies by means other than raising its wage differential. Finally m12 5 0 makes sense if it takes a larger increase of the firm's wage differen￾tial to facilitate the filling of some fraction of a given increment in itsvacancies the smaller is the unemployment pool from which workers can conven￾iently be drawn. The curve labeled m(u, z.) = 0 in Figure 1 gives the com￾binations of u and c that make A* = 0. Its slope, being -m2/ml, is necessarily positive, but the size of that slope and the curvature are in￾determinate and of no qualitative consequence. To the right of this locus A* > 0, and to the left A* < 0. In the United States and most other countries, satisfactory vacancy data are still unavailable. I shall couple the above model with a theory of labor turnover or employment dynamics, along lines suggested by Lipsey (l960), in ordertoderive testableimplicationsofrelations among easily observable data. The absolute time rate of increase of the aggregate number of persons employed, denoted N = dN/dt, consists of the number of persons hired l8 Stability of the average wage is consistent with some positive differentials if there exist firms content with negative ones. What counts for the average wage movement is the weighted average desired differential, A* (in relation to the ex post, actual, weighted average differential, say A, which necessarily equals zero)