Factor use and productivity change in the alcoholic beverage industries.

1. Introduction

Demands for wine, beer, and spirits are, in view of their substitutable nature, often studied in conjunction with one another (Heien and Pompelli 1989; Nelson 1999). Yet the alcoholic beverage industries are connected by similarities in their production processes and inputs as well as by the demand substitutability of their products (Lea and Piggott 1995). These industries are therefore usefully examined together when questions about technical change and productivity growth arise. Little recent attention has been given to the nature of technical adjustments in the beer sector, and no such work seems ever to have been published on wine and spirits. (1)

Brewing and distilling, and to a smaller extent winemaking, have become increasingly mechanized during the past half century as they have in most other manufacturing sectors. Packaging rates are higher, shelf lives longer, and storage and transportation technology improved (Tremblay 1987; Elzinga 1990, pp. 216, 219; Iserentant 1995; Piggott and Conner 1995; Gisser 1999). The fixed costs associated with these innovations have, broadly speaking, served as an incentive to build fewer and larger plants.

Industry concentration in the beer sector rose dramatically from the mid-1930s through the mid 1980s. Plant numbers then expanded as microbreweries catered to the new demand for product variety and idiosyncrasy, although four-firm concentration continued to rise through the mid-1990s. In the spirits sector, domiciled largely in the South, distillery numbers continue to decline, spurred by declining aggregate demand as well as scale economies. No such broad consolidation is taking place in the wine industry, where several California and New York firms long have occupied a substantial market share. Instead, wineries numbers have grown rapidly and uninterruptedly since 1970, proliferating in regions long thought inhospitable to wine grape production. Growing consumer interest in wine variety and microclimates, a trend similar to that for microbrews, has abetted this diversification (Goodhue et al. 2000).

Prices in the alcoholic beverage industries have, in real terms, fallen substantially in the past 40 years. Net prices to brewers have dropped 37%, to wineries 16%, and to distillers 31%. At the same time, real wage rates generally have risen, and, except for brief instability during the 1970s, so have capital rental prices. Real material prices in all three sectors rose from the 1950s through the 1960s, fell through the 1970s, then recovered by the mid-1990s to their early 1970s levels. Relative factor prices in the beer industry, broadly representative of those in the other two sectors, are shown in Figure 1. Wage rates rose relative to both capital and material prices until 1980. Thereafter, wages fell relative to capital but continued to rise relative to material prices (U.S. Bureau of the Census, various years).

Expenditure shares, illustrated in Figure 2, suggest that the beer industry is the most capital intensive of the three sectors. By the mid-1990s, capital expenditures represented one-third of beer production costs, compared to only one-fifth of wine and distilling costs. On the other hand, raw materials constituted little more than half of beer expenses but nearly 70% of wine and distilling costs. Labor accounted for only 10-12% of expenditures in any of these industries. Figure 2 seems to suggest that, at least since the mid-1970s or early 1980s, capital has substituted for labor and materials in all three sectors. Yet as Figure 1 reveals, much of these share changes are accounted for by the rising relative price of capital.

The substantial narrowing of output-input price margins in the alcoholic beverage sector suggests that the sector has become more cost efficient through either technical change, scale economies, improved utilization rates, or other means. Technical changes themselves can alter potential scale economies and productive capacities and hence prospects for future variations in industry structure. The exact form of the technical change also influences the distribution of factor demands, with implications for profitability in the machinery and farm production industries and for labor welfare. Finally, trends in final demand patterns, such as the rise of wine and beer niche markets, affect sector returns and thus incentives to expand output and improve efficiency. Disentangling these separate effects gives an improved view of the principal supply forces driving the alcoholic beverage sector. For example, a reliable characterization of factor substitution, which impinges on, among other things, input demands, requires we go beyond expenditure shares and examine the underlying processing technology.

To shed light on these issues, we investigate how the major features of beer, wine, and spirits production technology have changed during the past four decades. Our approach is to characterize the cost and input demand functions of representative firms, then employ duality principles to draw inferences for the underlying technological structures. The effort throughout is to distinguish among demand, technological, and scale influences on industry performance. To preview our results, we find productivity growth in all three industries to have been quite strong since the 1950s. Scale economies in brewing and distilling remain substantial, providing strong incentives for plant expansion. Virtually no scale economies, however, are evident in wine production. Substitutability among capital, labor, and material inputs is high, although technical change has shifted factor shares toward capital and away from materials.

2. Approach

Consider a firm's minimized cost

C = G(Y,[W.sub.l],[W.sub.m],K,D,t)+[W.sub.k]K],

where Y is output, K is the capital quantity, [W.sub.i] is labor wage, [W.sub.m] is materials price, [W.sub.k] is the rental price of capital, D is a dummy variable representing a discrete technology shift, and t is a time trend reflecting continuous technical change. L will refer to labor quantity, including skilled and unskilled workers, and M to material quantity, primarily raw farm products or prepared mashes, packaging supplies, and (to a minor extent) energy. Raw products in beer brewing are malted barley or other grain; in wine production, grapes; and in spirits, corn, rye, barley, and wheat (Lea and Piggott 1995). In 1992, the proportion of production materials accounted for by agricultural products was 13% in the beer sector, 50% in the wine sector, and 36% in the spirits sector. Most of the rest were in container supplies (U.S. Bureau of the Census, various years).

We employ the following generalized Leontief form for cost function G in Equation 1 (Morrison 1988, 1997; Park and Kwon 1995):

G = Y[([[alpha].sub.ll][W.sub.t]+2[[alpha].sub.lm][W.sup.0.5.sub.l][W.sup .0.5.sub.m]+[[alpha].sub.mm][W.sub.m])+([[beta].sub.ly][W.sub.l]Y.sup .0.5]+[[beta].sub.lt][W.sub.l][t.sup.0.5]+[[beta].sub.my][W.sub.m][Y. sup.0.5]+[[beta].sub.mt][W.sub.m][t.sup.0.5])

+[[beta].sub.yy]Y+2[[beta].sub.yt]Y.sup.0.5][t.sup.0.5]+[[beta].sub.t t]t)(W.sub.l]+[W.sub.m])]+[Y.sup.0.5][K.sup.0.5][)[[gamma].sub.lkd]D) [W.sub.l]+[[gamma].sub.lk]+[[delta].sub.lkd]D)[W.sub.l]+([[gamma].sub .mk]+[[delta].sub.mkd]D)[W.sub.m]

+ (([[gamma].sub.uk]+[[delta].sub.ukd]D)[Y.sup.0.5]+([[gamma].sub.yk]+[ [delta].sub.tkd]D)[t.sup.0.5])(W.sub.l]+[W.sub.m])]+([[gamma].sub.kk] +[[delta].sub.kkd]D)K([W.sub.l]+[W.sub.m]). (2)

Linear homogeneity in input prices and symmetry of the input-price Hessian matrix are imposed on this specification. Monotonicity in (Y, [W.sub.l], [W.sub.m],K), convexity in K (i.e., [[delta].sup.2]G/[[delta]K.sup.2] > 0), and concavity in factor prices are not and must be tested for.

Kerkvliet et al. (1998) and Gisser (1999) show that technical change in the 1970s beer industry was too sudden to be modeled adequately by a trend variable alone. Evidence suggests that similarly rapid changes were taking place during the same decade in the spirits sector and, perhaps to a smaller extent, in winemaking (Adams Business Media, annual issues; Lea and Piggott 1995). Goodwin and Brester (1995) find that the most pronounced structural changes in the aggregate food processing sector occurred in the 1980s rather than the 1970s, commencing in 1980 and largely completed around 1986. We examine the possibility of a relatively discrete technological jump during any of these periods by permitting the capital-related parameters [[gamma].sub.lk], [[gamma].sub.mk], [[gamma].sub.yk], and [[gamma].sub.kk] in Equation 2 to shift. (2) Alternative shift points were examined, including 1971, 1972, 1973, the early 1980s, and 1986. On the basis of goodness-of-fit criteria, a 1971 shift point was selected in all thre e industries. However, remaining parameter estimates were very robust to alternative shift points.

Output prices were specified as endogenous by modeling, for each representative firm, an output demand function of the form P = [[eta].sub.0] + [[eta].sub.1]Y + [[eta].sub.2]I + [[eta].sub.3]t, where P is output price and I is disposable personal income. The demands were estimated in conjunction with the firm's generalized supply function, [partial]G/[partial]Y = P[1 + [theta]/[[epsilon].sub.YP]], in which [[epsilon].sub.YP] [approximately equal to] P/[[eta].sub.1]Yis the output demand flexibility and [[theta] a market power parameter (Park and Kwon 1995). (3) Output pricing approaches the competitive norm as [theta] approaches zero and tends toward monopoly as [theta] approaches unity. Finally, labor and material demands were fitted by differentiating Equation 2 with respect to the corresponding factor price. Although nonlinear forms of the output demand functions might have been used, our data were found to be an unreliable basis for estimating own-price elasticities even in the vicinity of the sample mean. We therefore constrained [[eta].sub.1] to correspond at sample means to those in Heien and Pompelli (1989). (4) Estimates of the remaining parameters in Equation 2 were insensitive to parametric changes in these demand restrictions, suggesting little stochastic dependence between the system's demand and supply parameters.

Our principal interest is in identifying the sources of cost savings that have permitted price margins in the alcoholic beverage industries to narrow. In the long run--when the firm's productive capacity is fully utilized--two potential sources of such cost savings are technical improvements, associated with any downward shifts in the unit cost curves, and scale economies, associated with movement along any negatively sloped portions of the curves. The first is represented by the dual rate of technical change [[epsilon].sup.lr.sub.ct] = ([partial] ln C/[partial]t), namely, the elasticity of cost with respect to disembodied technical change as proxied by a time trend. The latter is often represented by [[epsilon].sup.lr.sub.cy] = [partial] ln C][partial] ln Y, the elasticity of cost with respect to output. It is straightforward to show that [[epsilon].sup.loc.sub.scale] = [[epsilon].sup.lr.sub.cy] -- 1 is the percentage change in unit cost associated with a 1% change in output. We will refer to [[epsilon].sup. loc.sub.scale] as the elasticity of (local) scale economy.

An additional source of cost savings, that from utilization improvements, is available in the short run to the extent the firm moves toward minimum-cost use of its fixed capacity, where capital's shadow price [Z.sub.k] = -- [partial]G/[partial]K and market price [W.sub.k] are equated with each other. Such utilization-based savings are confounded with the scale-based savings represented in the cost elasticity unless the latter is computed at the point where capital quantities K are adjusted to their long-run optimal levels K*. In that neighborhood, capital shadow and market prices are equal, so marginal output changes provide no incentive for capacity adjustments. (5)

The GL functional form permits solving explicitly for just such a long-run equilibrium capital quantity. Thus, it permits us to find the long-run dual technical change rate and cost elasticity, along with the long-run factor demand elasticities [[epsilon].sup.lr.sub.ij] = [partial]ln [X.sub.i]/ [partial]ln [W.sub.j], where [X.sub.i] is the ith input quantity and i, j = K, L, M. For example, since [W.sub.k] = [Z.sub.k] in equilibrium, the long-run dual technical change rate is [[epsilon].sup.lr.sub.ct] = [partial]ln C/[partial]t /K varying = [partial]ln G/[partial]t / [K.sup.0]=[K.sup.*]. Thus, obtaining [[epsilon].sup.lr.sub.ct] econometrically is a matter of differentiating Equation 2 with respect to t and evaluating the result at equilibrium capital quantity K*. Long-run marginal costs and long-run labor and material demands L* and M* are found similarly. The present study focuses on these long-run relationships.

To estimate these relationships for the three U.S. alcoholic beverage industries, we employ the SIC four-digit manufacturing data prepared by the Bureau of the Census and the National Bureau of Economic Research (NBER) (Bartelsman, Becker, and Gray 2000). Costs include production expenses only and exclude marketing and promotion costs. Data procedures, including those for formulating capital rental prices, are discussed in the Appendix. The sample series inns from 1958 through 1996. System estimation is conducted with 3SLS using SAS procedure SYSLIN. (6) Because expenditure share distributions and production processes in the three industries differ markedly from one another, we assume that the associated cost parameters differ also. The three industry models are therefore estimated separately. (7)

3. Productivity Growth and Scale Economies

Parameter estimates are reported in Table 1. Four of the five shift parameters in the beer model, two in the wine model, and four in the spirits model are statistically significant, suggesting that a discrete change did occur In these technologies in the 1970s. Nearly all cost parameter estimates in the beer and spirits models significantly differ from zero at the 5% level. Only about half of those in the wine model do so, possibly because of the wide product heterogeneity in this industry. Every regularity condition not maintained in the specification--monotonicity in factor prices, output, and capital; convexity in capital; and concavity in factor prices--was satisfied at every observation and in every industry. The suggestion is that the modeled factor uses are approximately cost minimizing. Morrison Paul (2001, pp. 532-3) also reports that the GL functional form has been generally successful in capturing regularity conditions.

In wine and spirits production, capital's shadow price - [partial]G/[partial]K exceeded market price [W.sub.k] from the mid-1960s to the mid-1980s, implying, consistent with the generally high demands during that period, that capital was too intensively utilized. Since then, and particularly in the spirits sector, capital has been underutilized. (8) Some utilization economies therefore were achieved in these two sectors in the late 1980s as capacities were brought into line with demands, although subsequent excess capacity brought corresponding utilization diseconomies. No such capacity utilization trends are evident in the beer sector, where capital's shadow price has oscillated in a narrow range around market price. During the 1990s in particular, capital use in the beer industry does not seem to have strayed far from equilibrium. On average over the 39-year period examined, capacity utilizations have approximately been optimal, providing some impetus to our focus on the long-run trends.

Productivity Growth

Dual technical change rates [[epsilon].sup.lr.sub.cr], that is, annual percentage cost savings induced by technical change, are shown together with cost-output elasticities [[epsilon].sup.lr.sub.cy] in Table 2. The dual rates have been highest in the beer sector, moderate in the spirits sector, and lowest in winemaking. Annual cost reductions from technical change have averaged 2.5% per year in beers, 1.7% per year in spirits, and 1.0% per year in wines. Low cost savings in the wine industry likely partly reflect the variety of wine types and microclimates, which, by introducing input fixities, discourage efforts to reorganize factors in a cost-reducing way. Higher savings in the beer industry are consistent with Tremblay's and others' findings that beer is a technologically dynamic sector. Nevertheless, the annual savings reported here are, by food industry standards, substantial in all three sectors. Morrison (1997) estimates that aggregate food manufacturing cost declined an average 0.73% per annum on acco unt of technical change between 1965 and 1991, and Chan-Kang, Buccola, and Kerkvliet (1999) give similar estimates.

A possible concern is that cost-reducing technical change rates have declined broadly over the years. Between 1961 and 1996, [[epsilon].sup.lr.sub.ct] dropped by one-half in the beer industry and by more than one-half in the wine and spirits industries. In the beer sector in particular, annual rates of cost reduction had flattened by the mid-1990s at 1.7% per annum. In wines and spirits, they had flattened to the vicinity of 0.5% and 0.9%, respectively. The suggestion is that automation and other rationalization efforts had gradually been "played out," reducing the potential for further efficiency gains.

On the other hand, our technology growth estimates likely are biased downward, especially in the beer and wine industries, by recent improvements in quality that have reduced output-input ratios at any given state of technology. Although difficult to model, such improvements are attested by the growth of specialty beer and wine products and by the growing product variety evident in the alcoholic beverage sections of most food stores. Quality and variety in many other food items have, of course, risen also, so more aggregate productivity measures probably are biased downward as well. Little tendency is evident in Table 2 of any recession-related technical change downturn. Studies at the aggregate U.S. food manufacturing (SIC 20) level, in contrast, have found productivity growth to often turn negative, particularly during recessions (Morrison 1997; Chang-Kang, Buccola, and Kerkvliet 1999).

Scale Economies

Cost elasticities with respect to output in Table 2 have remained below unity nearly every year in every industry. Hence, [epsilon].sup.loc.sub.scale], the proportionate unit cost savings that the sample-mean establishment would earn from a marginal output expansion, have largely been negative, implying locally increasing returns to scale. These local scale economies have been substantially higher in spirits and beer production than in winemaking. Indeed, wine production consistently has exhibited something close to constant returns to scale. Local scale economies in beer brewing rose through the early 1980s, then leveled off. By 1996, a 1% plant output increase would bring, at 1958-1996 sample-mean factor prices, a 0.29% decrease in beer's Unit production cost. Our finding of generally rising scale economies in brewing agrees with the results in Tremblay (1987) and in Kerkvliet et al. (1998); see also Elyasiani and Mehdian (1993). Scale economies in the spirits sector have behaved similarly to those in beer, rising through the early 1980s, then flattening out or declining. By the mid-1990s, they were about the same as those in beer production.

Local scale economies and cost-output elasticities each affect and are affected by the position and shape of unit cost curve, that is, by the scale economies that are globally achievable. Firms take advantage of global scale economies by boosting plant and firm size, actions that often exhaust the economies available. Table 3 provides insight into these relationships by comparing estimates of overall unit cost curve slopes with the corresponding mean unit costs, output prices, local scale economies, plant numbers, and mean plant sizes. To compute the unit cost curve slopes, we hold [W.sub.l], [W.sub.m], [W.sub.k], D, and t in Equation 2 fixed. Unit costs UC = C/Y then are generated at alternative output quantities Y and associated optimal inputs [L.sup.*](Y), [M.sup.*](Y), and [K.sup.*](Y), permitting capital quantity to adjust optimally to the given factor prices and output. Finally, Y is varied one and one-half standard deviations above and below the given year's output, and the mean unit cost curve slope i s computed from ([UC.sub.min] - [UC.sub.max])/([Y.sub.max] -- [Y.sub.min]) where min refers to the smallest and max to the largest value in the three-standard-deviation range. Input prices were held for this purpose at 1958-1996 constant-dollar (1994) sample means. (9)

Mean slopes of unit cost curves in the beer industry have remained rather constant since at least the mid-1970s, suggesting continued incentives for size expansion. And plant sizes did grow and plant numbers fall until around 1986. Total and per plant output growth, and the productivity-induced unit cost savings discussed previously, each served to boost local scale economies because proportionate changes in unit costs were being computed from increasingly smaller bases and proportionate output rises from increasingly larger bases (see Gisser 1999). This is the principal reason that local scale economies in the beer sector were rising until the mid-1980s since plant size growth along a given unit cost curve would otherwise have depleted such economies. That average brewery size now is declining in the face of constant or rising scale economies indicates that other factors, principally the public's new taste for specialty beers, lately have affected industry structure.

Unit cost curve slopes in the wine sector--never very substantial--have declined since the 1960s and for the past 15 years have been nearly flat. The concomitantly insignificant local scale economies in winemaking help explain why wineries have been able to proliferate in response to the rising demand for wine variety. That is, at least since the 1970s, wine demand growth largely has been accommodated by an increase in winery numbers rather than size, as one would expect in an industry in which costs are nearly constant and product variety is highly valued.

Global unit cost curve slopes in the spirits sector rose through the mid-1980s and have not declined much since then. Continued plant consolidation, seen in Table 3, was a natural consequence because little niche-product demand has been evident in liquors. Rising distillery sizes and productivity-induced unit cost declines combined, through their effects on the base points of proportionality changes, to boost local scale economies just as they did in the brewing sector. That is, cost-output elasticities fell despite the plant size growth, which otherwise would have pushed such elasticities upward. Only declining demand in the past decade has reversed this process by increasing the proportionate output changes represented by given output increments. Nevertheless, incentives for further distillery consolidation remain strong, as productivity growth continues to enhance the proportionate cost savings permitted by a given amount of plant expansion.

Overall, unit cost savings appear to have outstripped price declines in the wine and beer sectors. Mean per unit wine costs were 32% lower in 1996 than in 1966, while output prices dropped a mere 9% over the same period. Mean per unit costs in the beer sector fell about 52% from 1966 to 1996 compared to a real price fall of only 33%. Nevertheless, little evidence of market power can be discerned from these gaps since market power parameters 0 derived from Table 1 (0.08 in the beer sector, 0.01 in wines, and 0.16 in spirits) are quite low. The rising price/cost ratios instead suggest that consumers' willingness to pay for enhanced beer and wine quality, a concomitant of their higher incomes, has outstripped the cost of these enhancements. Gisser finds dynamic evidence that beer technology innovations in the 1970s and 1980s indeed reduced output prices, although concentration rates rose as a result. In the spirits sector, finally, costs have largely matched output prices: The cost curve minimum point shifted do wn 31% from 1966 to 1996, compared to a 27% drop in real output price.

4. Technology- and Price-Induced Factor Substitution

We have shown how technical changes in the alcoholic beverage sector have affected the impact of plant size and output on unit and total cost. A derivative question is how these cost impacts have been distributed among the chief factors of production. In general, inputs can be substituted for one another on account of changes in either technological possibilities or relative factor prices. Thus, before examining technological influences on factor deployment, it is informative to look at factor substitution in the price dimension alone.

Price-Induced Factor Substitution

Capital-intensive manufacturing often is associated with weak factor substitutability because the capital reconfigurations required when factor proportions are changed usually involve a substantial adjustment cost. Thus, the rising capital shares since the late 1970s depicted in Figure 2 lead one to suspect that factor substitution in the alcoholic beverage industries is weakening and that factor demands are becoming less elastic. This has not, however, generally been the case. In Table 4, we report own-price factor demand and Morishima substitution elasticities, evaluated at long-run equilibrium capital stocks at 10-year intervals during the 1958-1996 sample period.

In all three industries, own-price elasticities of labor demand have risen steadily and by the mid-1990s were near unity. Material demand elasticities have remained fairly constant and are substantially lower in the brewing than in the wine or distilling sectors. Own-price elasticities of capital demand are rather high in the wine sector, exceeding unity in every sample year. Overall, factor demands are most inelastic in brewing and least inelastic in winemaking, a fact explaining why beer cost shares (Figure 2) have been especially responsive to factor price changes while wine and spirit cost shares have stayed more uniform. These generally high factor demand elasticities imply a remarkable facility to substitute among production factors, a facility not popularly associated with production-line operations.

Morishima substitution elasticities, suggestive of factor substitutability at given technology and output, have with only one exception been positive in all three sectors. Substitutability in the wine industry is especially strong. Of particular interest are the rather high output- and technology-constant rates at which capital and materials substitute for one another ([M.sub.mk] and [M.sub.km]). In the wine and spirits sectors, these rates have exceeded even the substitutability between capital and labor. Capital intensification, in other words, has been highly responsive to relative capital and material prices, and the recent relative rise in the price of capital (Figure 1) presumably has significantly dampened capital's use. Substitutability between materials and labor has also been high, although in the beer sector, labor demands have fallen in response to material price increases (Mini < 0).

Factor Bias of Technical Change

Consider finally the factor cost share changes created through technology shifts. To do so, we must distinguish between shifts induced by technology itself and those induced by any associated scale changes. The percentage change in factor cost share created by technological growth alone is computed for the jth factor as

[[beta].sub.j] = [partial]In[S.sub.j]/[partial]t + ([partial]In[S.sub.j/[partial]In Y)(-[[epsilon].sub.ct]/[[epsilon].sub.cy]), j = L,M,K, (3)

where [S.sub.j] = [W.sub.j][X.sub.j]/C is the jth factor cost share. Biases [[beta].sub.j], j = L, M, K, together characterize the degree to which technical change shifts the factor expansion paths, altering the factor cost shares. Scale changes (represented by [partial] in Y) generally influence factor cost share changes also. The second right-hand term in Equation 3 corrects for such scale effects, leaving in the [[beta].sub.j] measures only those changes induced by expansion path shifts (Antle and Capalbo 1988, pp. 36-42). We permit capital to adjust optimally to t and Y, so our bias estimates are again long-run ones. (10)

Technical-change factor biases computed from Equation 3 are shown in Figure 3 for the entire 39-year sample period. In all three sectors, technical change has consistently been capital using ([[beta].sub.k] > 0) and material saving ([[beta].sub.m] < 0). Few trends seem evident in these biases other than that toward greater material saving in the beer sector. Holding cost constant, productive innovations in all sectors have substituted capital for raw materials. Expansion paths, in other words, have tended to shift toward capital and away from materials. This is illustrated in Figure 4, which depicts capital-material expansion paths at each middlecade in each industry. (11) The tendency of new technology to be generous with capital and conserving with materials has been especially prominent in winemaking. Interestingly, technical change in wine has been labor using, while in brewing and distilling it has, to a varying degree, been labor saving.

As Equation 3 shows, the cost-constant bias measures in Figure 3 consist of the gross expenditure share adjustments accompanying technical change (first right-hand expression) plus the scale effects of the change (second right-hand expression). In Table 5, we provide, for each of the three alcoholic beverage industries and each factor, the 1996 and mean 1958-1996 decompositions of technical bias [[beta].sub.j] into the gross expenditure share adjustment ([partial] In [S.sub.j]/[partial]t) and the scale-induced adjustment. Table 5 shows that scale effects in the wine and spirits industries have been small, suggesting that production functions in these industries have been nearly homothetic. In brewing, on the other hand, output growth along a given expansion path has strongly favored capital: Larger plants have been more capital intensive than have smaller ones. This effect has, in combination with the output-increasing effect of productivity growth, contributed strongly to the capital-using bias of technical change. Indeed, gross expenditure share adjustments [delta]In[S.sub.j]/[delta]t in the beer industry, where output and factor prices are held constant, have favored capital at an average rate of only 0.4% per year. But productivity-induced scale increases have accounted for an additional average rise of 1.9% per annum, so that the rise in capital cost shares created by expansion path shifts alone--evaluated along the equi-cost locus--has been 2.3% per annum. Scale effects on beer's labor and material shares have similarly been substantial.

As Table 5 indicates, technology change in wineries has induced an average annual 3.7% increase in capital's expenditure share and a 2.2% decline in material share, holding factor prices fixed and controlling for the scale effects of output growth. The mean bias against materials has been significant in the spirits sector as well, averaging 1.0% per annum. Because few changes have occurred in wine and liquor packaging, capital equipment in these two industries evidently has been selected for its ability to reduce raw material cull rates and thus the raw materials needed to yield a given quantity of final product. Bias against material use in brewing, however, has been slight. Given the low share of grain in beer material expenditures, this bias likely has been directed toward savings in packaging, such as in the adoption of aluminum-alloy cans, rather than savings in grain supplies (Gisser 1999).

5. Conclusions

During the past four decades, technical change in the alcoholic beverage industry has brought substantially higher multifactor productivity, lower unit costs, and lower real product prices. Productivity growth has been greatest in beer brewing and least in winemaking but in all three sectors has outstripped that in food manufacturing at large. Productivity growth has been continuous, defying both recessions and recovenes.

Local and global scale economies are substantial in the brewing and distilling sectors but, especially of late, weak to nonexistent in winemaking. Scale economies in beer brewing are rising, following a trend Tremblay identified 14 years ago. Those in distilling have only recently begun to decline, as plant sizes and industry volume contract in the face of receding demand. The approximately constant-cost nature of winemaking has varied little over the past four decades. Changes in these economies are driven less by trends in unit cost curve slopes than by productivity growth, which, by shifting unit cost curves downward, ensures that a given size-induced cost saving represents an ever larger percentage saving. Because aggregate incentives to expand enterprise size depend partly on local scale economies, plant expansion incentives appear to be stronger than ever in the brewing sector, weak in winemaking, and declining only slightly in the spirits sector. In the brewing industry, the increased demand for a micr obrewed products has acted as counterweight to expansion incentives.

Technical change in all three sectors has been capital using and material saving. Such technical substitution may until the late 1970s have been encouraged by declining relative capital prices. But the strong rise in capital charges since then suggests that innovation-related capital adoption has instead been stimulated exogenously, namely, by improvements in the quality of capital available to plant designers. These improvements arise from scientific and technological breakthroughs in the larger economy rather than from food processing itself. Amidst even the microbrew and niche-wine revolutions, capital-material expansion paths in the brewing and wine sectors are shifting toward capital.

Chief among our findings is a strong output- and technology-constant substitution between materials and capital. Buccola (2000) has shown in multiproduct assembly-line operations that the substitutability of material for nonmaterial inputs is a function of the rate at which firms substitute across product lines as relative input prices change. Thus, factor substitutability generally rises with the degree of data aggregation, and our conclusions apply strictly only to the SIC four-digit level of industry accounts. Keeping this caveat in mind, the results do suggest that any changes in relative factor prices would have a substantial effect on industry capital structure. In particular, further increases in relative capital prices will greatly discourage continued capital intensification in the absence of continued exogenous enhancements in capital quality. By the same token, the sector's capital demands likely will be highly sensitive to inflation, allowable depreciation rates, and other policy influences on imp licit capital taxation rates.

Appendix: Data Construction

Data in this study are taken from the National Bureau of Economic Research's (NBER) records for the beer (SIC 2082), wine (SIC 2084), and spints (SIC 2085) sectors, in conjunction with a Bureau of Labor Statistics' (2000) Multifactor Productivity Index series. NBER's four-digit data employ the 1972 SIC four-digit industry classifications rather than the 1987 definitions.

Output quantities are computed as the ratios of NBER value-of-shipments data to the respective shipment price deflators (1994 = 1). Quantities placed in inventory are excluded from such outputs because in the alcoholic beverage industries they may reasonably be regarded as yet-unfinished goods. Labor wage rates are determined as the ratio of total employee compensation to the number of hours worked by production and nonproduction employees. NBER's industry-specific material price deflators (1994 = 1) are used as the material price indexes. Dividing material price deflators into total material costs gives the corresponding material quantity indexes.

NBER data include capital quantities [K.sub.t] (weighted by base-year capital acquisition price [q.sub.k,0]) and capital acquisition price indexes ([q.sub.k,t]/[q.sub.k,0]) but do not include capital rental prices. The Bureau of Labor Statistics does report capital rental expenditures [W.sub.k,t][K.sub.t] at the SIC two-digit (food and kindred products) level. We allocated the latter to each of the three four-digit beverage industries according to that industry's proportionate share in the two-digit-level capital stock. Dividing such allocated expenditures by reported capital quantity [q.sup.i.sub.k,0][K.sup.i.sub.t] in the ith industry gives the industry's rental price as a percentage of base-year acquisition price. This method assumes that rental prices are, up to a multiplicative constant, identical in each four-digit food industry in a given year.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

Table 1

Cost and Factor Demand Parameters, Alcoholic Beverage Industries,
1958-1996

                                Beer                       Wine
     Parameter          Estimate  t-Statistie    Estimate  t-Statistic

  [[alpha].sub.lt]     0.095           3.58 *   0.026           1.35
  [[alpha].sub.Lm]     0.005           1.65     0.043           5.47 *
  [[alpha].sub.mm]     2.046          29.40 *   1.498          16.61 *
  [[beta].sub.ly]      0.001           4.39 *  -0.001          -2.71 *
  [[beta].sub.lt]     -0.034         -11.40 *   0.010           2.48 *
  [[beta].sub.my]     -0.002          -2.79 *   0.001           0.73
  [[beta].sub.mt]     -0.163         -15.09 *  -0.099          -6.05 *
  [[beta].sub.yy]     -0.000          -8.04 *  -0.000          -0.00
  [[beta].sub.yt]      0.000           9.82 *   0.000           1.52
  [[beta].sub.tt]     -0.000          -2.08 *  -0.001          -1.52
  [[gamma].sub.lk]    -0.076          -2.47 *  -0.025          -0.81
  [[gamma].sub.mk]    -0.690         -12.57 *  -0.830         -10.52 *
  [[gamma].sub.yk]    -0.001          -3.94 *   0.001           2.03 *
  [[gamma].sub.tk]     0.009           5.12 *  -0.11           -3.35 *
  [[gamma].sub.kk]     0.030           3.21 *   0.012           0.89
 [[delta].sub.LkD]    -0.040          -8.16 *  -0.013          -0.98
 [[delta].sub.mkD]     0.087           8.32 *  -0.113          -2.75 *
 [[delta].sub.ykD]     0.000           0.98    -0.000          -0.27
 [[delta].sub.tkD]     0.003           4.53 *  -0.001          -0.36
 [[delta].sub.kkD]     0.26           10.34 *   0.022           2.48 *
[[eta].sub.1][theta]  -0.245 E-4      -5.33 *  -0.128 E-4      -0.72
   [[eta].sub.0]       2.80           47.99 *   1.82           15.27 *
   [[eta].sub.1]      -0.294 E-3        --     -0.105 E-2        --
    [[eta].sub.2]      0.28 E-4        0.53     0.12 E-4        0.11
    [[eta].sub.3]      0.072           0.06 *   0.090           8.02 *
     [R.sup.2]         0.9957                   0.9963

                               Spirits
     Parameter          Estimate  t-Statistic

  [[alpha].sub.lt]     0.232           3.78 *
  [[alpha].sub.Lm]     0.038           7.71 *
  [[alpha].sub.mm]     2.882          32.43 *
  [[beta].sub.ly]     -0.001          -1.66
  [[beta].sub.lt]      0.001           0.22
  [[beta].sub.my]     -0.013         -10.76 *
  [[beta].sub.mt]     -0.112         -l7.96 *
  [[beta].sub.yy]     -0.000          -1.79
  [[beta].sub.yt]     -0.000          -2.26 *
  [[beta].sub.tt]      0.000           1.10
  [[gamma].sub.lk]    -0.568          -5.14 *
  [[gamma].sub.mk]    -2.372         -20.40 *
  [[gamma].sub.yk]     0.004           4.65 *
  [[gamma].sub.tk]    -0.005          -2.05 *
  [[gamma].sub.kk]     0.301           5.92 *
 [[delta].sub.LkD]    -0.059          -4.86 *
 [[delta].sub.mkD]    -0.003          -0.10
 [[delta].sub.ykD]     0.000           3.06 *
 [[delta].sub.tkD]     0.003           3.85 *
 [[delta].sub.kkD]     0.048           4.81 *
[[eta].sub.1][theta]  -0.126 E-3     -11.41 *
   [[eta].sub.0]       4.61           26.62 *
   [[eta].sub.1]      -0.802 E-3        --
    [[eta].sub.2      -0.81 E-3       -5.17 *
    [[eta].sub.3       0.083           5.09 *
     [R.sup.2]         0.9954

* Significant at 5% level

[[eta].sub.1] parameters were restricted (see footnote 4)

Table 2

Dual Productivity Growth Rates, Cost Elasticities and Scale Economies in
the Alcoholic Beverage Industries, 1958-1996

                                    Beer
           Dual           Cost              Scale
           Rate        Elasticity          Economiy
Year      [epsilon]ct  [[epsilon].sub.cy]  [epsilon].sub.cy] - 1

1961      -0.038         0.882              -0.118
1966      -0.030         0.888              -0.112
1971      -0.023         0.872              -0.128
1976      -0.021         0.838              -0.162
1981      -0.019         0.677              -0.323
1986      -0.019         0.757              -0.243
1991      -0.018         0.687              -0.313
1996      -0.017         0.714              -0.286
Mean      -0.025         0.777              -0.223

                                   Wine
           Dual           Cost             Scale
           Rate        Elasticity         Economy
Year      [epsilon]ct  [epsilon].sub.cy]  [epsilon].sub.cy] - 1

1961      -0.014         0.958            -0.042
1966      -0.010         0.962            -0.038
1971      -0.010         0.950            -0.050
1976      -0.010         0.968            -0.032
1981      -0.010         1.001            +0.001
1986      -0.007         0.975            -0.025
1991      -0.007         0.984            -0.016
1996      -0.004         0.964            -0.036
Mean      -0.010         0.971            -0.029

                                     Spirits
            Dual                Cost             Scale
            Rate             Elasticity         Economy
Year      [epsilon].sub.ct]  [epsilon].sub.cy]  [epsilon].sub.cy] - 1

1961       -0.026              0.908            -0.092
1966       -0.020              0.824            -0.176
1971       -0.016              0.717            -0.283
1976       -0.015              0.729            -0.271
1981       -0.016              0.547            -0.453
1986       -0.013              0.603            -0.397
1991       -0.010              0.689            -0.311
1996       -0.009              0.721            -0.279
Mean       -0.017              0.721            -0.279

Table 3

Global and Local Size Elasticities in the Alocholic Bevarage
Industries, 1996-1996

               Unit Cost  Mean Unit   Output          Local Scale
                 Curve      Cost     Price (b)          EConomy
Beverage/Year  Slope (a)    (C/Y)       (P)    ([epsilon].sub.cy] - 1)

Beer
  1966          -0.025      1.80       1.52             -0.112
  1976          -0.013      1.41       1.16             -0.162
  1986          -0.015      1.11       1.12             -0.243
  1996          -0.018      0.87       1.02             -0.286
    Mean        -0.018      1.30       1.21             -0.201
Wine
  1966          -0.066      1.47       1.11             -0.038
  1976          -0.029      1.21       0.96             -0.032
  1986          -0.008      1.08       1.01             +0.025
  1996          +0.004      1.00       1.01             -0.036
    Mean        -0.025      1.19       1.02             -0.020
Spirits
  1966          -0.069      1.72       1.39             -0.176
  1976          -0.086      1.43       0.98             -0.271
  1986          -0.098      1.21       0.89             -0.397
  1996          -0.081      1.19       1.02             -0.279
    Mean        -0.083      1.38       1.07             -0.281

                Number    Mean Plant
               of Plants     Size
Beverage/Year     (N)       (Y/N)

Beer
  1966            194       30.93
  1976            139       68.80
  1986            129       98.28
  1996            462       34.59
    Mean          231       58.15
Wine
  1966            209        5.33
  1976            257        8.28
  1986            479        6.74
  1996            660        6.78
    Mean          401        6.78
Spirits
  1966            110       23.74
  1976            108       29.67
  1986             79       44.81
  1996             61       42.40
    Mean           90       35.15

(a)Change in unit cost divided by change in output, where output is
varied plus and minus one and one-half standard deviations above and
below sample mean. Slopes are the indicated numbers divided by 1000

(b)Index (1994 = 1). All prices and price indexes are based on constant
1994 dollars.

Table 4

Long-Run Own-Price and Morishima Input Demand Elasticities, 1996-1996

                         Own-Price Elasticity
Beverage/Year  [[epsilon].sub.ij]  [[epsilon].sub.mm]

Beer
  1966               -0.390              -0.330
  1976               -0.191              -0.172
  1986               -0.419              -0.191
  1996               -1.067              -0.204
    Mean             -0.434              -0.236
Wine
  1966               -0.566              -0.614
  1976               -0.670              -0.820
  1986               -0.791              -0.796
  1996               -0.824              -0.641
    Mean             -0.706              -0.757
Spirits
  1966               -0.392              -0.619
  1976               -0.633              -0.633
  1986               -0.925              -0.653
  1996               -1.198              -0.621
    Mean             -0.716              -0.643

               Own-Price Elasticity    Morishima Substitution Elasticity
Beverage/Year  [[epsilon].sub.kk]  [M.sub.lm]   [M.sub.ml]  [M.sub.lk]

Beer
  1966               -1.027           0.194       -0.164      0.800
  1976               -0.608           0.141       -0.052      0.371
  1986               -0.594           0.340       -0.204      0.643
  1996               -0.722           0.906       -0.509      1.445
    Mean             -0.734           0.318       -0.174      0.715
Wine
  1966               -1.637           0.586       0.727       0.756
  1976               -1.116           0.840       1.708       0.593
  1986               -1.142           0.924       1.547       0.803
  1996               -1.272           0.883       0.982       0.983
    Mean             -1.303           0.824       1.396       0.732
Spirits
  1966               -0.574           0.486       1.022       0.389
  1976               -0.443           0.823       1.595       0.557
  1986               -0.448           1.090       1.705       0.906
  1996               -0.538           1.230       0.846       1.301
    Mean             -0.489           0.867       1.409       0.685

                 Morishima Substitution Elasticity
Beverage/Year  [M.sub.kl]   [M.sub.mk]  [M.sub.km]

Beer
  1966           1.911        0.946       1.552
  1976           1.023        0.600       0.831
  1986           1.408        0.562       0.864
  1996           2.502        0.547       1.087
    Mean         1.578        0.690       1.087
Wine
  1966           2.090        2.061       2.231
  1976           0.898        2.013       1.765
  1986           1.183        1.926       1.806
  1996           1.755        1.753       1.854
    Mean         1.371        2.035       1.942
Spirits
  1966           0.562        1.195       1.099
  1976           0.114        1.152       0.886
  1986           0.322        1.121       0.937
  1996           1.512        1.056       1.127
    Mean         0.438        1.163       0.980

Table 5

Decomposition of Long-Run Biases of Technological Change

                              Beer                       Wine
                  Gross     Scale      Tech     Gross     Scale
Years/Input      Effect    Effect      Bias    Effect    Effect

1958-1996 means
  Labor          -0.022    -0.029    -0.051     0.045    -0.008
  Materials       0.004    -0.008    -0.004    -0.025     0.003
  Capital         0.004     0.019     0.023     0.040    -0.003
1996
  Labor          -0.002    -0.042    -0.044     0.036    -0.004
  Materials       0.002    -0.012    -0.010    -0.017     0.002
  Capital         0.003     0.020     0.023     0.020    -0.002

                    Wine                 Spirits
                   Tech     Gross      Scale      Tech
Years/Input        Bias    Effect     Effect      Bias

1958-1996 means
  Labor           0.037    -0.025     -0.000    -0.025
  Materials      -0.022    -0.012      0.002    -0.010
  Capital         0.037     0.019     -0.002     0.017
1996
  Labor           0.032    -0.019      0.004    -0.015
  Materials      -0.015    -0.011      0.001    -0.010
  Capital         0.018     0.010     -0.001     0.009

Received June 2001; accepted October 2002

(1.) Gisser (1999) focuses on the relationships between a beer technology change proxy, output prices, concentration ratios, and welfare; he does not examine productivity growth itself. The most rccent work detailing productivity growth rates and scale economies in the beer sector appears to be in Tremblay (1987). Elyasiani and Mebdian (1993) and Kerkvliet et al. (1998) concentrate on technical inefficiencies, namely, those relative to a technical frontier.

(2.) Goodwin and Brester's procedure, taken from Bacon and Watts (1971). permits the shift in cost function parameters to start abruptly in a given year, then die out asymptotically. The method involves a grid search over alternative shift start years and decay rates and therefore is equivalent to our own procedure except in permitting shift decay rates to be noninstantaneous. Our decision to allow only capital-interaction parameters to shift discretely was guided by parsimony. Shifting all model terms would have involved estimating 30 parameters with 39 sample observations.

(3.) Product [[eta].sub.1][theta] is then estimated jointly, and [theta] is obtained through dividing [[eta].sub.1][theta] by [[eta].sub.1].

(4.) The restrictions were for beer (SIC 2082), -0.45; for wine (SIC 2084), -0.50; and for spirits (SIC 2085), -0.60. Gisser (1999) uses own-price beer demand elasticities of -0.90 and -1.24. Nelson (1999) estimates the beer, wine, and spirits demand elasticities to be -0.20, -0.69, and -0.11, respectively.

(5.) Morrison (1992, p. 385) shows this by decomposing the long-run cost elasticity with respect to output, [[epsilon].sup.lr.sub.cy], into the short-run (quasi-input-fixed) elasticity, [[epsilon].sub.cy], and the cost improvement obtained if the quasi-fixed inputs are permitted to adjust with Output.

(6.) We employed as instruments every variable in section 2 except for those that are endogenous by virtue of the equations auxiliary to Equation 2, namely, Y, P, L, and M. The exogenous cross products in Equation 2 were included in this set.

(7.) Panel data might have been employed along with shift terms permitting each parameter to vary by industry. However, this likely would introduce significant heteroskedasticity, especially considering that observed cost variances differ substantially across these industries.

(8.) By 1996, the total cost of distilled spirit production, computed at capital shadow prices, was just 85% of the cost computed at market prices. Inasmuch as capital's expenditure share at that time was only 20% of total Cost (see Figure 2c), there was a suggestion of extreme capital oversupply in the spirits industry in the mid-1990s.

(9.) These unit cost estimates become increasingly inaccurate as one moves away from each year's sample means. Estimates of the appropriate plant size ranges in this analysis are derived from aggregate data and hence suggestive only. Data on establishment numbers are collected at four-year intervals; those in nonreport years were estimated here by linear interpolation. Several establishments may be identified at a given plant or location if each is considered a separable activity (U.S. Bureau of the Census, various years). We use "establishment" and "plant" interchangeably.

(10.) Expression [partial] In [S.sub.j]/[partial] In Y in Equation 4 can be written as [partial] In [X.sub.j]/[partial] In Y - [[epsilon].sub.cy], where [X.sub.j] is the jth input quantity. Thus, long-run biases can be computed from the implicit function rule. Long-run derivative [partial][L.sup.*]/[partial]Y, for example, is found from [partial][L.sup.*]/[partial]Y = [partial][L.sup.*] + ([partial][L.sup.*]/[partial][K.sup.*]) ([partial][K.sup.*]/[partial]Y).

(11.) Only from the 1960s to 1970s in the beer and spirits industries do expansion paths in Figure 4 seem to shift in the direction of materials.

References

Adams Business Media. Annual issues. Adams Wine Handbook. New York: Adams Media Inc.

Antle, J. M., and S. M. Capalbo. 1988. An introduction to recent developments in production theory and productivity measurement. In Agricultural productivity: Measurement and explanation, edited by S. M. Capalbo and J. M. Antic. Washington, DC: Resources for the Future, pp. 17-95.

Bacon, D., and D. G. Watts. 1971. Estimation of the transition between two intersecting straight lines. Biometrika 58:525-34.

Bartelsman, E. J., R. A. Becker, and W. B. Gray. 2000. NBER-CES/Census Manufacturing Industry Productivity Database. Accessed 9 December 2002. Available http:/www.nber.org/nberces/nbprod 96.htm.

Buccola, Steven. 2000. Materials and value-adding inputs in manufacturing enterprises. Journal of Productivity Analysis 13:227-43.

Chan-Kang, Connie, Steven Buccola, and Joe Kerkvliet. 1999. Investment and productivity in Canadian and U.S. food manufacturing. Canadian Journal of Agricultural Economics 4:105-18.

Elyasiani, Elyas, and Seyed Mehdian. 1993. Measuring technical and scale inefficiencies in the beer industry: Nonparametric and parametric evidence. Quarterly Review of Economics and Finance 33(4):383-408.

Elzinga, Kenneth G. 1990. The beer industry. In The structure of American industry, edited by W. Adams. New York: Macmillan, pp. 203-39.

Gisser, Micha. 1999. Dynamic gains and static losses in oligopoly: Evidence from the beer industry. Economic Inquiry 37:554-75.

Goodhue, Rachael E., Dale M. Heien, Hyunok Lee, and Daniel A. Sumner. 2000. Contracts, quality, and industrialization in agriculture: Hypotheses and empirical analysis of the California winegrape industry. Unpublished paper, University of California, Davis.

Goodwin, Barry K., and Gary W. Brester. 1995. Structural change in factor demand relationships in the U.S. food and kindred products industry. American Journal of Agricultural Economics 77:69-79.

Heien, Dale, and Gregory Pompelli. 1989. The demand for alcoholic beverages: Economic and demographic effects. Southern Economic Journal 55:759-70.

Iserentan, D. 1995. Beers: Recent technological innovations in brewing. In Fermented beverage production, edited by A. G. H. Lea and J. R. Piggott. London: Blackie Academic and Professional, pp. 45-65.

Kerkvliet, Joe R., William Nebesky, Carol Horton Tremblay, and Victor J. Tremblay. 1998. Efficiency and technological change in the U.S. brewing industry. Journal of Productivity Analysis 10:271-88.

Lea, A. G. H., and J. R. Piggott. 1995. Fermented beverage production. London: Blackie Academic and Professional.

Morrison, Catherine J. 1988. Quasi-fixed inputs in U.S. and Japanese manufacturing: A generalized Leontief, restricted cost function approach. Review of Economics and Statistics 70:275-87.

Morrison, Catherine J. 1992. Unraveling the productivity slowdown in the United States, Canada, and Japan: The effects of subequilibrium, scale economies, and markups. Review of Economics and Statistics 74:381-93.

Morrison, Catherine J. 1997. Structural change, capital investment, and productivity in the food processing industry. American Journal of Agricultural Economics 79(1):1 10-25.

Morrison Paul, Catherine J. 2001. Cost economies and market power: The case of the U.S. meat packing industry. Review of Economics and Statistics 83:531-40.

Nelson, Jon P. 1999. Broadcast advertising and U.S. demand for alcoholic beverages. Southern Economic Journal 65:774-90.

Park, Seung-Rok, and Jene K. Kwon. 1995. Rapid economic growth with increasing retums to scale and little or no productivity growth. Review of Economics and Statistics 77:332-51.

Piggott, J. R., and J. M. Conner. 1995. Whiskies. In Fermented beverage production, edited by A. G. H. Lea and J. R. Piggott. London: Blackie Academic and Professional, pp. 247-74.

Tremblay, Victor J. 1987. Scale economies, technological change, and firm cost asymmetries in the U.S. brewing industry. Quarterly Review of Economics and Business 27(2):71-86.

U.S. Bureau of the Census. Various years. Census of Manufactures. Washington, DC: U.S. Department of Commerce.

U.S. Bureau of Labor Statistics. 2000. Major sector multifactor productivity index. Washington, DC: U.S. Department of Labor.

Yin Xia *

Steven Buccola +

* Department of Agricultural Economics, 124 Mumford Hall, University of Missouri, Columbia, MO 65211; USA; E-mail [e-mail]; corresponding author.

+ Department of Agricultural and Resource Economics, 213 Ballard Hall, Oregon State University, Corvallis, OR 9733 13601, USA; E-mail [e-mail].

Support for this research has been provided by the National Research Initiative Competitive Grants Program, CSREES, U.S. Department of Agriculture. The authors thank Victor Tremblay, Department of Economics, Oregon State University, for valuable comments on an earlier draft.