PRODUCTION ORGANIZATION AND__ EFFICIENCY DURING TRANSITION

Author(s): Erik Mathijs and Johan F. M. Swinnen

Source: The Review of Economics and Statistics, Vol. 83, No. 1 (Feb., 2001), pp. 100-107Published by: The MIT Press

Stable URL: http://www.jstor.org/stable/26693Accessed: 19/09/2009 03:58

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at

http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/action/showPublisher?publisherCode=mitpress.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with thescholarly community to preserve their work and the materials they rely upon, and to build a common research platform thatpromotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Review ofEconomics and Statistics.

http://www.jstor.org

PRODUCTION ORGANIZATION AND EFFICIENCY DURING TRANSITION: AN EMPIRICAL ANALYSIS OF EAST GERMAN AGRICULTURE Erik Mathijs and Johan F. M. Swinnen* Abstract-Enterprise restructuring is expected to improve efficiency in transition economies. With data from former East Germany, we compare the efficiency of family farms and partnerships with large-scale successor organizations of the collective and state farms (LSOs). Using parametric and nonparamnetric techniques, we show that LSOs display lower technical but that this difference is efficiency than do family farms and partnerships small and declining during transition, mainly as a result of structural changes in agriculture. Family farms are not as scale efficient as partner- ships and LSOs, and partnerships are superior to all other organizational forms. Jozef Konings, Karen Macours, Erik Schokkaert, and two anonymous referees for helpful comments. The views expressed in the paper are those of the authors and do not necessarily reflect those of the European Commission. 1 See, for example, the special issue of the Journal of Comparative Economics edited by Ben-Ner, Brada, and Neuberger (1993) for a com- prehensive overview of the issues, and Brada, King, and Ma (1997) for an analysis of enterprise efficiency in Hungary and the Czech Republic. functional form of the production function. Using nonpara- metric techniques (which does not require the choice of a specific functional form and which allows decomposition of technical efficiency into a pure technical effect and a scale effect), Piesse, Thirtle, and Turk (1996) find that private Slovenian dairy farms displayed higher technical efficiency but lower scale efficiency than did dairy cooperatives be- tween 1974 and 1990. This paper presents the first dynamic analysis of differ- I. Introduction ences in technical and scale efficiency between farm orga- n important issue in central and eastern European nizations after the 19 reforms in CEE countries, using methods.2 East German (CEE) transition is the restructuring of production both parametric and nonparametric organizations and its impact on productive efficiency. A key data are used because German statistics are the most con- question is whether so-called de novo firms are more effi- sistent and reliable of that presently available in former former state enterprises, when both communist economies. Section II presents the hypotheses cient than restructured are facing hard budget constraints in a competitive environ- on the efficiency of different organizational forms in agri- ment.1 In agriculture, these de novo firms are typically culture. The data are discussed in section III. Efficiency of family farms. Former measures are calculated in section IV (parametric method) family farms or small partnerships collective and state farms that have not been liquidated have and in section V (nonparametric). Section VI discusses the typically been transformed into either \"private coopera- results, and section VII concludes the paper. tives\" or shareholder companies (Swinnen, Buckwell, & Mathijs, 1997). We refer to both organizations as large- II. Production Organization and Efficiency scale successor organizations (LSOs). in Agriculture Empirical evidence on the relative efficiency of such farm A farm is technically efficient if it maximizes output with organizations is scarce and limited to comparisons during earlier reforms in East Asian socialist countries and to given inputs after having chosen technology. A farm is also combination corresponds to pre-reform comparisons in those CEE countries where fam- scale efficient if its input-output during communism. The shift from a zero-profit, long-run competitive equilibrium situation, ily farms were important collective farms to individual farms had a large positive that is, when it produces at constant returns to scale (Fare, impact on efficiency and growth in China (McMillan, Whal- Grosskopf, & Lovell, 1985; Chavas & Aliber, 1993). The ley, & Zhu, 19; Lin, 1992) and in Vietnam (Pingali & organization and governance of the farm affects its technical Xuan, 1992). Pre-19 comparisons of small-scale family and scale efficiency. In this paper, we test four hypotheses farms and large-scale collective (or state) farms in Poland on efficiency differences between farm organizations in and Yugoslavia show no difference in technical efficiency transition countries. between state farms and family farms (Boyd, 1987, 1988). Later studies using frontier estimates confirm Boyd's results A. Hypothesis 1: Family Farmns Have Higher Technical for Poland (Brada & King, 1993), but not for Yugoslavia LSOs. Efficiency Than (Hofler & Payne, 1993). These studies use parametric esti- Family farms are expected to display higher technical mation techniques, which makes their results sensitive to the efficiency than LSOs because of the LSOs' inherent princi- Received for publication May 16, 1997. Revision accepted for publica- pal-agent problems in labor contracting due to the difficulty tion February 9, 2000. of linking effort in production to income (Schmitt, 1991, * Katholieke Universiteit Leuven, and Katholieke Universiteit Leuven 1993; Pollak, 1985). Metering effort in production is par- and European Commission, respectively. We thank the Fund for Scientific Research-Flanders (Belgium) for ticularly stringent in agriculture because of its biological financial support. We are grateful to Volker Beckmann, Konrad Hagedorn, and sequential nature and spatial dimensions (Brewster, 2 Many qualitative and theoretical analyses are available. For former East Germany, see, for example, the debate between Peter and Weikard (1993) and Beckmann, Schmitt, and Schulz-Greve (1993) on the relative efficiency of production organizations. Beckmann and Hagedom (1997) provide a more general overview of structural change in East German agriculture. The Reviewv of Economiics and Statistics, February 2001, 83(1): 100-107 Institute of Technology and Fellows of Harvard o 2001 by the President College and the Massachusetts AN ANALYSIS OF FARM PRODUCTION ORGANIZATION DURING TRANSITION 101 1950; Binswanger & Rosenzweig, 1986). Family farms are argued to be more efficient than LSOs in this regard because family members maximize family welfare rather than indi- vidual welfare, and hence face no incentive to \"free ride.\" are lower (Carter, Therefore, the costs of monitoring effort 1984). B. Hypothesis 2: The Efficiency Gap between Family Farms and LSOs Falls During Transition. TABLE 1 -AVERAGE FARM SIZE IN HECTARES AGRICULTURAL LAND AND SHARE IN TOTAL AGRICULTURAL LAND OF DIFFERENT ORGANIZATIONS IN EAST GERMANY, 1992-1995 1992 Family farms average size share Partnerships average size share Cooperatives average size share Companies average size share 1993 1994 1995 State and collective farms performed poorly not only Source: Beckmann and Hagedom (1997). 46 ha 45 ha 48 ha 46 ha 13% 18% 20% 21 % 626 ha 511 ha 469 ha 449 ha 14% 18% 21 % 22% 1,537 ha 1,265 ha 1,218 ha 1,1 ha 44 % 36 % 34 % 39 % 2 ha 823 ha 772 ha 1,006 ha 24 % 23 % 29 % 26 % because of their intrinsic problems, but also because of extrinsic problems, such as bureaucratic controls and an extractive external environment (Johnson, 1983; Putterman, 1985; Lin, 1990; Brada & King, 1993). As transition progresses, these controls are removed, and the environment is liberalized; the efficiency of the LSOs will improve. For example, LSOs can solve their agency problems by setting up the right labor contract structure. The difficulty of mon- itoring in an agricultural production cooperative can be overcome by self-enforcing contracts, in which members promise to discipline themselves (Lin, 1990). Such a self- enforcing contract can be sustained only in a free coopera- tive-in which each member has the right to exit-and is therefore more likely to be effective in a restructured orga- nization and a liberalized environment.3 C. Hypothesis 3: The Efficiency Gap between Family Farmns and LSOs Depends on the Specialization of the Farms. Problems of governance and worker supervision are fewer if activities can be easily monitored in terms of inputs or outputs (for example, if work gangs can be organized and supervised directly, or if output can be measured directly, so that workers can be paid on a piece-rate basis, (as in harvesting) (Pollak, 1985). Furthermore, labor-intensive ac- tivities are more sensitive to monitoring problems, ceteris paribus. Therefore, large livestock farms may display lower technical efficiency than large crop farms because linking output to labor effort is generally more difficult in livestock production than it is in crop production, and because live- stock production is, on average, more labor intensive than arable crop production. D. Hypothesis 4: There Are No Scale Economies in Agriculture, Except for the Small Family Farms That Have Lower Scale Efficiency. Scale economies would favor the largest organizations. However, there are no increasing returns to size in agricul- tural production beyond a certain minimum size which can 3 maximum Each member effort or to shirk, decides either to honor his commitment to provide disintegration. raising the possibility of the cooperative's their contract and The threat of dissolution ensures that exert maximum effort. members commit to be captured by (larger) family farms, both in developing (Berry & Cline, 1979; Hayami & Ruttan, 1985) and in developed countries (Kislev & Peterson, 1991; Peterson, 1997). The long-run, average-cost curve in agricultural production is L-shaped under long-run competitive equilib- rium conditions (Hallam, 1991). III. Data The data are from official German statistics. Agricultural output, Y, is production value in German marks (DM) deflated by the consumer price index (OECD, 1996). LAND is cultivated land in hectares (ha). LABOR is the number of annual work units, and CAPITAL is the sum of the value of buildings, machinery, and livestock in DM deflated by the consumer price index.4 The available data are averages of panel data from 1991- 1992 to 1994-1995. In 1994-1995, this panel consisted of 729 sole proprietorships or family farms, 137 partnerships, and 301 shareholder companies (including cooperatives) or LSOs (Agrarbericht, 1996). For family farms and partner- ships, data are available for two farm specializations (crops and livestock, based on the share of livestock in total revenues), and further for three size categories for family farms (according to standard farm income). LSOs are cat- egorized into three farm specializations (crops, livestock, and mixed). As a result, eleven aggregate observations (two for partnerships, three for LSOs, and six for family farms) are available for each year, and the data set has a total of 44 observations. The composition of the panel changed with the restruc- turing of the farms. Table 1 displays the relative importance of each of the organizational forms and their average size in terms of cultivated land in East Germany during transition. Table 2 summarizes the descriptive statistics of the sample used for the analysis. Family farms or \"sole proprietors\" are farms worked and managed by a single household. They vary from less than 40,000 DM to more than 100,000 DM standard farm in- come. Their share of total cultivated land increased from 4 may be sensitive Because output and capital are expressed in monetary terms, the LSOs in how assets to differences between are valued and how they react family farms, to policy changes. partnerships, results and 102 THE REVIEW OF ECONOMICS AND STATISTICS DATA IN THE SAMPLE FOR 1991-1992 AND 1994-1995 OF ENTERPRISE TABLE 2.-DESCRIPTIVE STATISTICS Output (thousand DM) 1991-1992 Family farms Crops Livestock Partnerships Land (ha) 1991-1992 161 65 425 180 2,017 1,716 1994-1995 220 91 594 2 2,041 1,716 Labor (work units) 1991-1992 1.9 1.7 3.9 4.1 51.8 80.0 1994-1995 2.2 2.0 5.6 4.3 40.0 47.5 DM) Capital (thousand 1991-1992 345 333 851 710 4,326 6,768 1994-1995 483 501 967 1,171 4,812 5,474 1994-1995 406 234 1,179 748 5,228 5,162 313 197 944 576 4,970 6,167 Crops Livestock LSOs Crops Livestock Source: Own calculations based on Agrarbericht (1996). 13% in 1992 to 21% in 1995, and their average size varied between 45 ha and 48 ha. In the sample, the average size of these farms is considerably higher: 199 ha for crop farms and 81 ha for farms that specialized in animal products. On average, family farms in the sample employ between 1.4 and 2.8 labor units. Partnerships are new forms of collaboration, mostly a few persons (they employ approximately five labor units on average) who are related to one another (Beckmann & Hagedorn, 1997). Their share of total cultivated land in- creased from 14% in 1992 to 22% in 1995. On average, the in the sample have 250 ha of land in livestock partnerships production and 534 ha in crop production, which is similar to the overall averages. LSOs include cooperatives and joint stock companies, both characterized by limited liability. Most of these farms are transformed collective farms. By 1992, 40% of the collectives had been liquidated, and 30% were transformed into cooperatives, 25% into companies, and 8% into part- nerships (Beckmann & Hagedorn, 1997). Although cooper- atives are generally larger than joint stock companies, the data do not allow disaggregation between them, such that any conclusions will apply to LSOs in general. The share of LSOs in total cultivated land decreased from 73% in 1992 to 57% in 1995. By 1995, the LSOs in the sample used around 2,000 ha and fifty labor units on average, which is consid- erably more than the East German average. Using output value, Y, as a dependent variable yields biased efficiency calculations if the output value of different organizations is differently affected by price changes. How- ever, the differences in product mix and output prices are are relatively small. Although family farms and partnerships somewhat more specialized than LSOs are, output price differences between organizations were small (less than 5%) in 1994-1995. In 1991-1992, price differences were some- what larger, especially for wheat (10% to 15%). The price differences have decreased during transition, as market transparency increased. Interestingly, there is no simple correlation between the 1991-1992 price differences and farm organization: LSO crop farms received the highest wheat prices, but LSO livestock farms received the lowest milk prices. IV. Parametric Estimation of Efficiency Efficiency is calculated by comparing the observed out- put of a certain activity with the maximum output using the same inputs.5 The maximum or \"efficient\" output level is located on the frontier production function. The same fron- tier is used for crop, mixed, and livestock farms, because most farms in our sample engaged in both crop and live- stock activities. Following Lin (1992), Hofler and Payne (1993), and Piesse, Thirtle, and Turk (1996), we use a Cobb-Douglas specification to estimate the production function frontier, because the data do not allow the estimation of more- sophisticated functional forms.6 The production function frontier, which can be either stochastic or deterministic, can be written as + &3T93/94 + &4T94/95 ln (Y) = (l + Ce2T92/93 + PA ln (LAND) + L ln (LABOR) -u, (1) + PK ln (CAPITAL) + where &( denotes the intercept for 1991-1992, Ti is a dummy variable for year i (Ti = 1 for year i and 0 for other years), and is the production elasticity at the frontier for input j. f3j The time dummies are included to account for yearly shifts of the frontier.7 The error term is decomposed into a two- sided error term that is normally distributed, v - N(O, o-a), and represents statistical noise, and a one-sided error term, iu ' 0, that follows a half-normal distribution with mean R and variance o2 and represents technical inefficiency. The stochastic frontier is estimated using the maximum-likeli- hood technique of LIMDEP following Aigner, Lovell, and Schmidt (1977). S Activities can be different farms at a certain point of time, a particular farm in time, or a combination of both. at different points 6 The estimates thus derived are the Timmer measures of technical efficiency (Timmer, 1971). 7 Dummy variables for each year were used instead of a time trend to reach convergence in the maximum-likelihood estimation. AN ANALYSIS OF FARM PRODUCTION ORCGANIZATION DURING TRANSI'TION 103 TABLE 3.---PARAMFTRIC FRONTIER ESTIMATION WITH LN(Y AS DEPENDENT VARIABLE Deterministic frontier with time without time Stochastic frontier dummies dummies Intercept 1.04 (0.6) 1.46 (1.7)* 2.81 (2.8)*** Time duniinies T92/93 - 0.19 (-5.2) * - -0 12 (-34)*** T93/94 --0.22 (-6.2)*** --0.16 (-4.3)*** T94/95 --0.24 (--5.9)*** -0.17 (-4.6)*** Iiputs Ln(LAND) 0.67 (19.4)*** 0.62 (16.5)*** 0.63 (13.2)**! Ln(LABOR) -0.12 (-1.2) --0.07 (--1.4) 0.02 (0.3) Ln(CAPITAL) 0.66 (5.0)*** 0.63 (8.6)*** 0.51 (5.9)*** SuIn 1.21 1.18 11.6 Adjusted R2 0.997 0995 # observations 44 44 44 t-values in parentheses. Statistical significance is indicated at the 1% (**s), 5% (*), and 10% (*) level. Excluding statistical noise from specification (1) yields the deterministic frontier: In (Y) = a I+ a2T9293 + a3T93/94 + a41 94/95 + PA In (LAND) + 1L In (LABOR) (2) + K In (CAPITAL) - wlhere all coefficients and variables have the same interpre- tation as in (I). The error terni contains only a one-sided negative error term, ui. We estinmated the deterministic frontier using corrected ordinary least squares (COLS) (Greene, 1980; Russel & Young, 1983; Piesse, Thirtle, anid Turk, 1996). It involves first the OLS estimxiation of the average production function, ln(Y), which is characterized by an intercept a1: In (Y) = a1 + a2T9?293 + CX3T93/94 X- a4T94/95 + PA In (LAND) +- 1L In (LABOR) (3) + O3K In (CAPITAL) + v where v denotes statistical noise. The observation with the largest positive etror vmax is considered the most efficient observation. To yield the frontier, the average production function is shifted upwards, such that the new intercept & I =- a + vnx, and that the observation with the largest positive residual lies on the frontier ln(Y) - -11 + a2T92/93 4- a3T93/94 + a4T94/95 -+- PA In (LAND) + f3, In (LABOR) (4) + Kln (CAPITAL). Trhis approach assumes that the slope parameters of the average production function and the fronitier are the same. The results of both methods are summarized in table 3. The results are similar, which is not surprising with average data. The estimated coefficients in both estimations are close. All coeffici.ents are statistLically significant at the 1% level, except for the intercept and the coefficient jor labor. Two observations indicate the possibility of a inisspeci- fication with the Cobb-Douglas function. First, the coeffi- cient of labor :is nec ative, or positive but close to 0 when the time dummnies are not incluided (third column of table 3). Second, the sum of coefficie-nts is equal to 1.21 and 1.18 for the stochlastic and the deterministic frontiers, respectively, indicating that increasing returs to scale are the doninating feature of the observations in the sample. Introdutcing a slope dumrmly i) equation (3) for LSOs oni the coefficient for LAND considerably increases the coefficient for labor (from -0.07 to 0.35), while decreasing the other coeffi- cients. TFhe sumi of coefticieits is then, close to I for LSOs, while it is still larger than I for family farms. This result indicates that there mxay exist mor-e than one frontier and that farmily farms may display increasing retu:rns to scale, while LSOs display constant retuirns to scale. However, one should be caireful with the interpretation of regressions using slope dumnies. As pointed out by Brada and King (1993), the estimated coefficients may reflect differences in input use rather than differenices in production technology. Pooling, tests based otn asymptotic likelihood compparisonis of the restricted (that is, without slope dum- mies) and unrestricted (with slope dlummies) specification support pooling of the d'ata for fa nily farns and partner- ships and for LSOs: the vidual slope variables (5.24 for land, 3.86 for labor, 4.20 for X.2 statistics for i itroducing indi- czapital, and 4.30 for the intercept) all lie between the critical ;X2 values at the 95% and 99% confidence levels (3.84 and 6.63, respectively). These tests support the view of Brada and Kiig. T'herefore, we calculate the efficiency measures for individual observatiois based on a restricted specifica- tion. Nevertheless, because the sample is doninated by fanily farms and partnerships (32 of the 44 observations), estimationis may be biased towards family farins and part- terships, such1 that the efficier cy measures for LSOs may be underestimated. To calculate the technical efficiency measures for indi- vidual observations, we used the determ7inistic frontier with- out the time dumimies, such that shiifts of the frontier during transition now show up in the efficiency estimnationIs. Tech- nical efficiency measures for tdie- individual observations are calcultate d bv subtractitio the largest positive residual from all residuals setting the error term of the rmost efficient c)bservation equal to 0. Exponentiation gives the relative efficiency of each observation as a percentage of the most efficient one. Th'be results of these calcul-ations (table 4) suggest that, on average, partierships had the higlhest technical efficiency, but the diff.erence with family fwrms is small. At the begin- ning oif transition, both partnerships anid family farms were, on average, approximiately 210% mi:-ore efficient, than LSOs. This efficiency gap was significantly reduced duiing tran- sition. By 1994-4995, the efficiency galp was reduced to 104 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 4.-PARAMETRIC MEASURES OF TECHNICAL EFFICIENCY Crops Livestock 1991-1992 1994-1995 1991-1992 1994-1995 Famnily farmns 85.9 76.9 96.1 76.0 Partnerships 87.3 82.0 100.0 78.9 LSOs 70.9 70.5 76.7 79.1 LSOs/family farms 82.5 91.7 79.8 104.1 LSOs/partnerships 81.2 86.0 76.7 100.3 Source: Own calculationis. lO%15% for crop farms and had ceased altogether for -ivestock faris., As discussed aRbove, these results may be affected by our methodology (in particular the choice of the Cobb-Douglas model). results fuither, we analyze Therefore, betfore interpreting and analyzing the whether the results are robust to changes in the methodology. In order not to impose a priori restrictions, flexible functional forms can be used. However, because we have no a priori information on the actual fornm, a method that does inot rely on strict parametric assumptions is preferred (Chavas & Cox, also because it allows calculation of both 1995). For this reasoni-and technical and scale efficiency measures-we use nonparametric techniques to calculate efficiency measures in the next section. V. Non-Parametric Estimation of Efficiency Estimating a nonparametric deterministic frontier, ex- pressed in terms of minimizing input requirements, allows the decomposition of a measure of \"total technical effi- ciency\" (ET) into a measure of \"scale efficiency\" (Es) and of \"pure technical efficiency\" (EP), that is, the nonparamet- ric Farrell measure of technical efficiency.8 As in Fire, Grosskopf, and Lovell (1985), we assume that productioni is characterized by a nonparametric, piecewise- linear technology, so that simple linear progranmming tech- niques can be used. We fturther assume strong disposability of ouitputs and inputs. For each activity k, total technical efficiencv ET, can be estinmated using the following linear program, allowing only for constant rettuns to scale: minA,Z A S t. ZY Y Yk ZX AXk (5) where Yk denotes the output of activity k, Xk is a vector of inputs employed by activity k, and z is a vector of k intensities that characterize each activity. To calculate the pure technical efficiency mileasure, Er, the restriction that the sum of intensities be equal to 1 is added to the previous 8 See Farrell (1957). This technique is also called data envelopmnent analysis (DEA). Onjly radial measuires of technical efficiency are calcu- lated. One should keep in mintid equiproportionate that such measures allow for only outputs. reduction of inputs or equiproportioniate expansion of progran to allow for constatnt and nonconstant returns to scale: mnin,k, A S.t. ZY > Yk zX!5 AXk z?>O (6) k Zk lo Scale efficiency can then be calculated by dividing total efficiency by pure technical efficiency, Ef = ET/Ef. Fi- nally, to identify the source of scale technical efficiency (E*) is calculated inefficiency, the star- for the scale-ineffi- cient activities using the following linear program that allows for only nonincreasing returns to scale: mlnA,z X S.t. ZY ?Yk ZX?AXk (7) Z ~2 0 k E Zk 1 i=l An activity exhiibits increasing returns to scale if E* < and decreasing returns to scale if E- = activity, by definition, exhibits constaint ET. A scale-efficient Et returns to scale. To compare with the parametric calculations, the non- parametric measures are calculated using the whole data set rather than the observations of the same year. Note that a comparison of parametric and nonparametric efficiency measures can be done in only relative terms, because the reference point differs: the frontier is defined differently for the Cobb-Douglas specification than it is for the nonpara- metric method. The results are in table 5. To test whether the differences in efficiency are signifi- cant, the efficiency measures were used as dependent vari- ables in a Tobit regression analysis (because the efficiency measures are censored at 1). The explanatory variables are in table 6. The variable LSO is 1 for LSOs and 0 otherwise; PARTNERSHIP is 1 for partnerships and 0 otherwise; TREND takes the values 0, 1, 2, and 3 for the respective years; and CROPS is 1 for crop farms and 0 otherwise. VL Results and Discussion Several conclusions can be drawn from the results in tables 5 and 6. First, there is no statistically significant difference in (pure) technical efficiency between family farms and LSOs for crop falrns. In livestock production, family farmis lhad higher pure technical efficiency at the beginning of transition: the difference is 13% in 1991-1992. However, this gap has disappeared during transition. 1994-1995, there is no longer any significant difference By AN ANALYSIS OF FARM PRODUCTION ORGANIZATION DURING TRANSITION 105 TABLE 5.-NONPARAMETRIC MEASURES OF PURE TECHNICAL, SCALE, AND TOTAL TECHNICAL EFFICIENCY Crops Livestock 1991/ 1994/ 1991/ 1994/ 1992 1995 1992 1995 Pure technical efficiency Famnily farms 100.0 97.8 100.0 93.4 Partnerships 100.0 96.9 100.0 100.( LSOs 96.9 100.0 87.2 94.9 LSOs/fanily farms 96.9 102.2 87.2 101.6 LSOs/partnerships 96.9 103.2 87.2 94.9 Scale efficiency Family farmns 82.2 82.3 87.2 87.8 Partnerships 100.0 99.9 99.9 99.8 LSOs 96.3 92.7 100.0 99.9 LSOs/family farms 117.2 112.6 114.7 113.8 LSOs/partnerships 96.3 92.8 100.1 100.1 Total technical efficiency Family farms 82.2 80.5 87.2 82.0 Partnerships 100.0 96.8 99.9 99.8 LSOs 93.3 92.7 87.2 94.8 LSOs/family farms 113.5 115.2 110.7 115.6 LSOs/partnerships 93.3 95.8 87.3 95.0 Source: Own calculations. between family farms and LSOs for livestock, as for crops. These results are consistent with hypotheses 1 through 3 (that family farns were technically more efficient than LSOs, but only in certain specializations (such as livestock) and only in the beginning of transition). With institutional reform, organizational the extemal restructuring, and liberalization of environment, the gap in efficiency has disap- peared during transition for all specializations. The im- provements in technical efficiency and governance of the LSOs is also reflected by the labor adjustments in LSOs. Employment was substantially reduced on all LSOs, but especially on livestock LSOs, where average emnployment decreased by 45% over the 1991-1995 period from 80 to 47.5 labor units. TABLE 6.-TOBIT REGRESSION ON NONPARAMETRIC EFFIcIENCY MEASURES Total Pure technical Scale technical efficiency efficiency efficiency Intercept 100.9 85.6 83.9 LSO (5166.1)*** --4.3 (J480.4)*** 19.8 (12.6)*** 13.2 PARTNERSHIP (37)** 4.2 (14,6)** (9 19.5 20.3 1) (4.0)** (26.3)*** (36.0)*** TREND -1.8 -0.3 -1.7 (9.4)*** (0.1) (2.8)* LSO X T'REND 2.5 -1.2 1.8 (5.4)** (0.3) (0.8) LSO X CROPS 1.0 -9.7 -5.9 (0.2) (4.6)** (1.8) Log likelihood -97.8 -135.3 -141.5 # of observations 44 44 44 Chi-square values in parentheses. Statistical significanlce is indicated at the 1% (***), 5% (**), and 10% (*) level. lower scale efficiency Second, family farms have, on than both partnerships average, significantly and LSOs. In livestock production, there is no difference in scale effi- ciency between partnerships and LSOs, although in crop production the largest fanrs (LSOs) have lower scale effi- ciency than do patrtnerships. These results are conisistent with hypothesis 4 (that there are no positive scale effects beyond a certain minimum size). The results indicate that, on that average, family farms are below this minimum size and partnerships are at or above this minimum size. Fur- thermore, LSOs in crop production seem to be too large; that is, they reflect negative scale effects. Third, the combined results indicate that partnerships were the most efficient organizational formr in forner East German agriculture during transition. Partnerships combine high levels of pure technical efficiency due to good gover- nance of labor (often family relatives and small in number) and full economies of scale by operating on sizes than average family farms. larger farm Fourth, the TREAND and TREND X LSO coefficients indicate that there is a decline in the average technical efficiency of partnerships and especially of famfily farms during transition, and that this effect is offset for the whole sample by a significant increase in the technical efficiency of the LSOs. The dynamics of farm restructuring in East Germany may explain this decreasing trend in technical efficiency for family farms and partnerships. Because the data used in our analysis measure onlv the average technical efficiency of each type of farm, tle restructuring process changes the average technical efficiency of both LSOs and family farms. Individual workers will leave the LSO to start up a family farnm if they expect to benefit from this; the larger the benefits, the more likely their departure (Matl-hijs & Swinnen, 1998). Therefore, the first individuals to leave the LSOs were those with the highest expected income gain, determined both by their expected labor individual family farming and productivity in by the labor productivity of the LSO. As a consequence, the worst-perforMing LSOs were first liquidated because workers were more likely to leave them (or because they went bankrupt). This liquida- tion process improved the average efficiency of the remain- ing LSOs, in addition to increasing the technical edue to irnproved m,anagement and restructuring of the re- fficiency maininig LSOs. Furthermore, the first individuals to leave LSOs are those who expect to have the highest productivity as individual farmers. Hence, when the next group leaves, the average labor productivity of family farms may decrease, reinforc- ing the decline in the average labor productivity gap be- tween LSOs and family farns. This transition restructuring process is consistent with the results in table 6, as the share of famnily farms and partnerships of total land use increased from 27% to 43% between 1992 and 1995. (See table I.) Because output is measured in monetary terms, differ- ences in producer prices combined with differences in 106 AND STATISTICS TIHE REVIEW OF ECO'NOMICS product mix across farm types may affect the results. LSOs and liberalization of the external. environment, the gap in are somewhat less sPecialized than are family fairms and effiiency has disappeared during transition tor all special- partnerships, anid, fr-om 1991-41992 to 19944-1995, wheat izations. The improvements in techn1ical efficiency and gov- prices decreased and mrilk prices increased. Therefore, be- erance of the LSOs are also reflected in strong reductions cause crops in general-and wheat in paricular.--make tu.p irn labor use, especially in livestock LSOs. Our results are no increasing returms to scale beyond a smaller share of total reventies in). crop LSOs (67% to 73% confirm that thiere anid a certain mininimum farmus crop famnaily crops and 40% to 45% wheat) than in. size, which is captured by partnerships 50% to 55% wheat), but not by the average faniily farms. LSOs ini crop produc- partnerships (85% to 90% crops antd are the crop LSOs were less affected by the decrease in wheat tion exhibit decreasing returns to scale. Partnerships prices. Vice versa, it may also have contributed to the most efficient or-anization combining high levels of pure persistent lower technical efficiency of livestock LSOs conm- technical efficiency due to good labor governance with low pared to partnerships in 1994-4995. At the samie time, employment, often relatives, and full economies of scale by increased milk prices patily explain vihy thze efficiency of operating on lafyger than average family farms. The farmzis livestock LSOs has increased more -than that of crop LSOs. analysis indicates a declirne in the average technical effi- crop farmi) holds in general: avertage ia-tter Tl.he effiTiency ciency of family farms and partnerships during transition. declined while livestock :arni efficiency increased. [iis is due to a combination of the effects of fann restruc- Finally, regarding the methodology, the parametrc atnd turing on the average efficiency level of pooled data, and calculations are consistent; for the most- also to different output mnixes and price effects of the fanr noonparametric results. Both approaches organizations. important anrd mnost-significant fax ms was efficienicy of faymtiy indicate that pure techinical initially higher than it was in LSOs, that the effici lncy gap REFERENCES that this decline is most impor- declined durin-g transitioni, tant for livestock, and that there is a negative trend in. the Agrarbericht ,Agrar- und errnihrungspolitischer Bericht der Bundes- regierung (Bonn: Bundesministerium fir Erndhirung, Land- mecasures paramnetric average efficiency mneasures. The non wirtschaft und Forsten, 1993, 1996). between were higher on average, and showed less differen)ce Aigner, Deinnis, C. A. Knox Lovell, and Peter Schmidt, \"Formulation and meassures. This is organizations than did the pararrmetric Estinmation of Stochastic Frontier Production Function Models,\" Journal method compares individuua1 of Econometrics 6(1) (July 1977), 21-37. because, firstly, the parametric \"Decollectivisationi and Priva- Konrad Hiagedorn, observations to the single most-efficient observations and Beckmann, Volker, anid Changes of Agriculture in tisation Policies and Resultirig Structural with (several\") ore- the nonpararnetric techi-ique comwpares in Johani F. M. Swinnen, Allan Buckwell, and Eastern Germiany,\" Sec- Erik Mathijs (Eds.), Agricultural Privatisation, Land Refb,-nz and efficient observations witth a comparable inpuit mi-ix. Farm Restructuring in Centi-al and Eastern Europe (Aldershot: convex hull iietric procedure forms -lhe ondly, the nonpara Ashgate, 1997), 105-160. of the observations, more closely embracing the data thant Beclukmanni, Volke; GUInther Schmitt aid Willi Schulz-Greve, \"Diskussions- does the pw ametric frontier. In addition, the calculation of beitriige: Betriebsgofie und Organizationsform fUir die land- Agramwirtschqvi 42(11) wii schaftliche Produktion. Anmnerkuntgen,\" both pure techniical and scale efficiency is an irmportant (1993), 412-419. Ini summnary, Beni-Ner, Avner, Josef C. Brada, and Egon Neuberge r (Eds.), \"The advantage of the nonparanetric maethodology. of efficienlcy n 1asures re- because parametric estimnation] and Behabvior of Econonuc Orgarnizations: Theoretical StructuLre Economics and Emnpirical Persp ectives,\" Jlournal of Conmparative quires a priori specPiication of a functional for n an.d be- 17(2) (1993), 201.-206. cause it does not allow decomposi.tion of effficiency inito BerTy, anid Production R. Albert, anid Williarn R. Cline, Agrarianz Structure efficiency cal- efficiency, nornp