The effect of technology transfer program participation on small farms in Chile.
Christopher M. Edmonds
International Rice Research Institute
Los Baños, Laguna Province, The Philippines
Selected paper for presentation at the Annual Meetings of the American Agricultural Economics
Association, August 8-11, 1999, in Nashville, Tennessee.
Address where correspondence should be sent:
International Rice Research Institute
P.O. Box 3127
Makati Central Post Office
1271 Makati City
This research examines the effect of a government sponsored technology transfer program for
small holder farms in Chile. The effect of family participation in the technology transfer program is
evaluated with respect to a number of 'outcomes' including its effect on farm revenues, total family
income, and household poverty status. The empirical examination uses maximum likelihood selection
and fixed- and random-effects estimation techniques. By estimating the effect of program participation
on crop selection, crop yields, farm use of certified seeds, and the scale of farming activities, the
research examines the mechanisms through which the program appears to raise farm revenues.
Estimation results show program participation had a positive and significant effect on farm revenues
and total family income. The program prompted farmers to adopt nitrogen fixing bean crops, but did
not have significant effects on crop yields or the likelihood a farm planted certified seeds or applied
fertilizer. The primary mechanism through which the program increased farm income was by
increasing the intensive scale of farming pursued by participants.
Keywords: Agricultural Extension Services, Program Evaluation, Impact Assessment, Technological
Change: Choices and Consequences, Chile, Farm Households, Panel Data.
The author acknowledges support from a dissertation field research fellowship from the Joint
Committee on Latin American Studies and an International Pre-dissertation fellowship of the Social
Science Research Council and the American Council of Learned Societies, with funds provided by the
Ford Foundation. Support was also provided by the Fulbright Commission and a travel grant from the
Center for Latin American Studies, University of California, Berkeley. The Giannini Foundation for
Agricultural Economics provided assistance during the initial phases of the research.
Copyright 1999 by Christopher Edmonds. All rights reserved. Readers may make verbatim copies of
this document for non-commercial purposes by any means, provided that this copyright notice
appears on all such copies.
1. Introduction and overview.
This paper examines the performance of a Chilean government sponsored technology transfer
program (the Programa de Transferencía Technológica (PTT) in Spanish) for small farms using data
from a longitudinal survey of households in a single southern province in Chile. The effect of the
program, and the role of farm characteristics (i.e. farm size, labor endowment) on farm performance is
evaluated using a number of economic welfare and agricultural performance measures. Estimates of
farm and total family income serve as measures of the final outcome of PTT participation. Estimates of
the scale of farming, crop adoption and yields, and technology use corroborate results from income
estimates and indicate how changes in income were attained.
Information concerning the effect of agricultural technology transfer programs in developing
economies is of interest to both government policy makers and international donors seeking to allocate
limited funds for rural development. The technology transfer program in Chile has been characterized
as a model of semi-privatized agricultural extension (e.g., Carney, 1998, Bebbington and Sotomayor,
1998). This makes examination of its effectiveness of broader interest.
To date, examinations of the effect of technology transfer on small farms in Chile have mainly
been qualitative in nature (e.g., Falaha-Lumi, 1992, Sotomayor, 1994). Lopez (1994) quantitatively
examined the technology transfer program using data from a region that overlaps the area from which
data for the present study is drawn. This study found that the PTT had a positive but not statistically
significant effect on family income and the level of farm output. The estimation techniques applied in
this study did not control for the self-selection of households into the program.
2. The technology transfer program for Chilean small holders.
Agricultural technical training and technology transfer services offered through the PTT are
available to farming households with small land holdings and few assets. To participate in the PTT a
family can own no more than 12 irrigated equivalent hectares of land and have assets valued no more
than 3,500 UF—roughly equivalent to $87,500 US in 1995. The program has the broad object-ive of
improving the agricultural practices applied on farms and ultimately enhancing farm income. The
technology transfer program is typical of policy measures pursued under Chile’s "Growth with Equity"
development strategy which seek to alleviate poverty and foster the integration of groups at the margin
of the commercial economy within the framework of a liberalized and market-driven economy. The
program’s self-financing components and use of private-public partnerships are also typical of this
Agricultural technical assistance to small farms in Chile has been provided in some form since
the 1960's.1 The present program dates back to technical assistance programs provided to small-tomedium-sized farms beginning in 1978. In the 1980s, the government moved to public financing
agricultural extension contracted through private technology transfer organizations. The number of
farms participating in the PTT program and expenditures for the program have increased markedly
since the mid-1980's (see Edmonds, 1998, for details). In each community served, PTT activities
occur through committees of about twenty families from the locality. Committees meet about six times
a year, and training and other program activities are conducted during meetings.
3. Modeling the effect of the technology transfer on agricultural households.
The decision to participate in the PTT program is modeled as resulting from a household's
assessment of their welfare as participants in the program compared with the level of welfare they
obtain if they do not participate in PTT. Households participate in the PTT because they determine
their utility as participants is higher than the level of utility they would obtain as non-participants. The
costs and benefits of PTT participation depend, in turn, on many of the same characteristics that
determine a household's income. A farm's benefit from program participation depends upon the need
the farm operators have for the training and services provided by the program, and the farm’s capacity
to apply the technologies introduced in the PTT. Farm adoption of PTT technology, in turn, depends
upon the amount of land a family owns, its capital holdings, and similar characteristics. The cost of
participation is primarily the lost labor time and costs associated with transport to program meetings,
and charges for specific goods and services provided. Unmeasured household characteristics such as
ambition and social capital will bear heavily on the benefits derived from the program. Survey data
cannot accurately quantify such household characteristics. This makes it essential that statistical
techniques be employed that control for such unobservable characteristics.
The estimation proceeds from the structural equations for farm income:
Y Farm = ∑ Y j = ∑ p j Ψ ( Lt ) F j ( L j , H j , T j , K j ; Z )
Farm income is a function of prices ( p j ), the quantity of each of k crops produced by the farm,
labor input from family (L) and hired non-family (H) sources, the amount of land cultivated (T), and
the amount of capital (K) applied to the production of each crop j, a technical efficiency parameter (Ψ)
defining the farm productivity which depends upon the amount of family labor time devoted to PTT
activities (Lt), and other exogenous farm characteristics (Z). We assume productive inputs and outputs
are non-joint. The expression values agricultural output at market prices. The values of goods used
only for home consumption are imputed using the price of the nearest neighbor that reported selling the
Estimation of equation (1) directly raises a number of problems. Variable input levels are
likely endogenous with the level of farm production. Estimating equation (1) would require estimation
of each of K crops separately, while available data only provides information on aggregate levels of the
Several works in Spanish give detailed descriptions of the programs to assist Chilean peasant farms
and review the changes the PTT has undergone since its inception (Berdegué, 1994, Leiva and
inputs per farm. Treating PTT participation as an exogenous variable leads to omitted variables bias
due to self-selection of families into the program.
While levels of labor, land, and capital actually applied by households in farming are likely
endogenous with farm income, a household's endowments of these inputs can be considered
predetermined. Farm production of k distinct crops is collapsed into a single production function
defining the value of agricultural output of the farm. Data do not permit identification of the labor time
devoted to PTT activities (Lt) so program participation is reduced to a single dummy variable (D) that
takes on a value of one if the household participated in the PTT program and a zero otherwise. With
these changes, expression (1) simplifies to:
Y Farm = ΨD ⋅ F ( L, T , K , Z )
For simplicity, we assume the production technology is characterized by a Cobb-Douglas
production function. This functional form captures the expected concavity between inputs and the level
of output. It is adequate for the present purpose since our intention is to develop an expression for
farm income that is amenable to estimation: Y Farm = ΨDL β1 T β 2 K β 3 Z 1β 4 Z 2β5 ⋅ ⋅ ⋅ Z Oβ O
Finally, we take the natural log of both sides of expression (3) making it linear in logs.
4. Treatment effect estimation procedures.
The principal statistical problem faced when estimating the effect of program participation
when households can self-select into the program is the potential for omitted variables bias. If program
participants have unobserved characteristics which are correlated with their decision to participate in
PTT and these are also correlated with farm income (or other outcome measures), estimates of
program effect computed from the estimated coefficient on the dummy variable defining household
participation status will be biased. We apply two estimation techniques to account for the effect of
self-selection of families into the PTT in assessing the effect of participation in the program: 1. a fixed-
effects estimator, and 2. a random-effects estimator.2 The alternative estimates allow us to check the
sensitivity of results to the specification and to compare results with those of previous studies. The
random-effects estimator includes a household-specific error term, and the fixed-effects estimator
includes a household-specific intercept term to characterize the effect of unobserved household
characteristics on program outcomes. A disadvantage of the random-effects estimator is that it
requires the assumption that individual effects are uncorrelated with other regressors. The Hausman
test examines the validity of this assumption. We report the Hausman specification test statistics in all
panel data based estimates. When household specific-effects are included, the estimated coefficient
and asymptotic t-test for the dummy variable indicating farm participation in the PTT program will
provide an unbiased measure of the impact of the program—provided the unobserved differences
across households are constant over time and can be characterized as household-specific errors in the
5. Overview of the survey area and data examined in estimations.
The data for the present study were collected in Ñuble Province, which is part of the Eighth
Region in central-southern Chile. Data come from a random sample of roughly two hundred households in the Province. The survey was administered in 1987 to collect information on farm/ house-hold
characteristics and the agricultural activities pursued during the 1986-87 agricultural year (July 1 to
June 30). The follow-up survey was conducted in 1995, collecting information for 1994-95.
Of the 208 households identified in 1986-87 survey: 176 completed the follow-up survey, 21
were known to have abandoned the farm they owned or rented in 1986-87, two refused to be interviewed, and 16 households could not be located. These 16 households also probably left their farms, but
this could not be confirmed from interviews with former neighbors. Some information on the 21
The maximum likelihood form of a Heckman two-step estimator for estimation of the program
participation and farm income outcomes as endogenous variables using cross-sectional data from each
families known to have left their farms was collected from interviews with their former neighbors.
Among the 174 households that completed both the initial and the follow-up survey, more than thirty
percent took part in the PTT each year. Table 1 summarizes data from the survey.
Table 1. Summary statistics on variables used in evaluation of the effect of the PTT.
Assoc. of sugar enterprise
Household head’s age
Household head’s schooling
Pct. income from agriculture
Own some type of transport
Extreme poverty (Cash)
Extreme poverty (Total)
Regular poverty (Cash)
Regular poverty (Total)
Use of certified wheat seed
Source: Nuble Panel Survey.
Ratio (children/adults) 0.377 0.479
Fulltime equiv.mnths. 15.566 7.280
Irrig. equivalent has.
Pesos (x 100,000)
-Pesos (x 100,000)
Pesos (x 100,000)
0/1 (cash inc./capita)
-0/1 (total inc./capita)
-0/1 (cash inc./capita)
-0/1 (total inc./capita)
-Quintiles per ha.
6. Estimation results.
Estimates of the effect of farm participation in PTT on farm income are reported on Table 2.
We discuss fixed-effects estimates, but they are not reported. Random- and fixed-effects estimates of
farm income were similar. The fixed-effects panel model explained more of the variance in farm
income. The Hausman test statistic supports use of the random-effects estimator. Estimates show the
PTT participation had a positive and statistically significant effect on farm income. PTT participation
caused a 47.5 percent increase in household farm income according to the random-effects estimator.
year of survey was also estimated but is not reported due to space.
Table 2.Summary of program effect estimates: Farm and total family Income, and poverty status.
(Estimation Standard Error)
Time dummy variable
Participant in technology transfer program (0 or 1)
Associate of quasi-public sugar enterprise (0 or 1)
Age of head of household
Years of education of household head
Family dependency ratio (children/adults)
Household labor endowment (months avail. For work)
Land endowment (irrigated equivalent hectares owned)
Value of farm equipment/machinery held by household
Household owns some type of private transport (0 or 1)
Distance to the Provincial capital
12 dummy variables identifying locality of farm
Goodness of fit measures and other diagnostics
Log-Likelihood Ratio Test statistic X2 [11 d.f.]
Lagrange Multiplier Test [2 d.f.]
Panel--Hausman test (fixed vs. random effects) [9 d.f.]
Probability Hausman test statistic
Notes: n.r. (not reported for ease of exposition).
/1 None of the coefficients estimated for the localities
*** statistically significant at α=.01
were statistically significant in 1986-87.
** statistically significant at α=.05
/2 Asymptotic standard error reported.
* statistically significant at α=.10
The fixed-effects estimator attributed a 43.5 percent increase in farm income from PTT participation.
In terms of the effect of participation on the level of farm income, results imply participants increased
their farm incomes by 59,000 pesos in the random-effects and by 52,000 pesos in the fixed-effects
estimators (roughly equivalent to $193 and $130 US, respectively).3 These increases were close to the
average cost of the PTT (about $150 US, MIDEPLAN, 1991).
Estimated equations showed a positive and statistically significant effect of farm association
with the quasi-governmental sugar enterprise on farm income. The magnitude of this increase was
greater than that obtained for the PTT. Other variables with statistically significant effects on farm
income were: labor endowment, amount of land owned, and ownership of private transportation.
The fourth column of Table 2 summarizes estimates of the effect of PTT participation on total
family income. Results are similar to those of farm income estimates—participation in the PTT had a
positive and statistically significant effect on total family income. The Hausman test fails to reject the
null hypothesis that individual effects are uncorrelated with other regressors.
Approximately sixteen percent of the households surveyed in 1986-87 attired from the panel.
This raises possible attrition biased in results. Two models (selection models using 1986-87 data with
selection on attrition) were estimated to test for this. Results suggest attrition bias is not present.
We present estimates of the effect of PTT participation on other outcome measures (in order):
poverty status, farm adoption of selected crops, crop yields, use of certified seeds, and the intensive
scale of farming activity. Based on estimation results just discussed, we employ only panel data
estimators and set aside concern about attrition bias. Considering these additional outcomes provides
an opportunity to verify the results obtained from the estimates of farm income. Estimates of the effect
of PTT participation on farm crop selection, yields, technology applied, and scale of farming activities
To approximate the implied effect of the coefficient in terms of the change in the level of the left hand
also provide insight into the relative importance productivity increases, changes in cropping systems,
and increases in the intensity of farming activity in farm income increases.
Having found farm participation in the PTT had a positive and statistically significant effect on
farm and total family income, a remaining question is whether the PTT was successful in assisting the
needy households. This can be considered by examining the effect of PTT participation on the poverty
incidence using a probit model with a random-effects error structure (Butler and Moffit, 1982).4 The
monthly per capita income levels defining the 'extreme' and 'general' rural poverty lines are compared
to per capita monthly income to determine household poverty status.5 Estimates were highly
statistically significant according to goodness of fit measures. The random-effects model correctly
predicted the poverty status of 75 percent of the households. As shown on Table 2, participation in the
PTT was found to reduce the probability that the household's per capita income was below the poverty
line, but the estimated coefficient was not statistically significant. This result was obtained regardless
of whether the estimation considered the general or extreme poverty line, or cash or imputed income.
There was a statistically significant decline in poverty incidence among surveyed farms over time.
Other variables found to have statistically significant effects on poverty were: association with the
sugar enterprise, the age of the household head, and the size of a family's landholding. With the
income increases brought to PTT participant farms, these results imply that the program has not been
successful in targeting its services.
side variable, we apply the conversion: β i ( level ) = e
( βˆi − 1 Var ( βˆi ))
− 1 (Kennedy, 1981).
For completeness, a fixed-effect model based on the logit distribution proposed by Chamberlain
(1984) was also estimated, but is not reported.
6.1 Non-income measures of the effect of PTT participation.
PTT participation was estimated to have a positive and statistically significant effect on the
number of hectares planted. The estimated marginal effect of PTT participation on the land cultivated
was an increase in the planted area by 33.7 percent or 0.39 hectare. Other variables having significant
effects on cultivated area were: sugar enterprise association, the household dependency ratio, labor
endowment, and ownership of transport and capital equipment.
Farm use of certified high yielding varieties of wheat seed provides an indicator of the technical
level of the farms and success of the PTT in disseminating new technologies. According to the
estimate, PTT participation increased the likelihood of certified seed use, but the effect was not
statistically significant. Certified seed use among surveyed farms increased over time, and the trend
was statistically significant. Association with the sugar enterprise was estimated to increase the
likelihood of certified seed use significantly. Older farm operators were significantly less likely to use
certified seeds. Larger farms and farms with more capital were more likely to use certified seed.
The PTT encourages farms to adopt new crops in place of wheat and other traditional crops.
Table 3 shows PTT participation had a positive and statistically significant effect on the likelihood of
vegetable cultivation. Farms in the PTT were 13.1 percent more likely to cultivate vegetables.
Estimates show PTT participation increased wheat yields. The estimated coefficient on the
dummy variable defining PTT participation was significant at a .06 level of significance. On average,
PTT participation raised the wheat yield by 24 kilos. There was a trend toward increasing yields
between 1986-87 and 1994-95. The only other right hand side variable estimated to have a significant
effect on wheat yields was farm association with the sugar enterprise.
The 'extreme' rural poverty line represents the cost of purchasing a bundle of goods required to
maintain a nutritionally adequate diet. We used Chilean rural poverty lines from 1993, deflating the
values to their equivalent value in 1986-87 and 1994-95 using the Chilean consumer price index.
Table 3. Estimates of program effect on other outcome measures.
(Estimation Standard Error)
Time Dummy Variable
Sugar enterprise associate
Age of head of household
Household head education
Value of farm capital
Private transport owned
----Goodness of fit measures: R2
Log-likelih. ratio test [11 d.f.]
Lagrange Multip. Test [1 d.f.]
Hausman test [9 d.f.]
/1 Asymptotic standard error.
Actual 0 39 37
1 23 149
7. Conclusions and implications of the research.
This research provides evidence that the technology transfer program for small farms in Chile
had a positive and statistically significant effect on participants' incomes. Results control for selfselection of households into the PTT and are consistent across alternative specifications of the
estimation equation. The estimated increase in income accruing to families participating in the
program was more than the average expenditure per farm on the program. The positive effect of
program participation was obtained using a number of outcome measures. The effect of program
participation on household poverty status was generally not statistically significant, suggesting program
services need to be better targeted to the most needy households.
Examination of intermediate program outcomes such as crop adoption, use of particular
agricultural technologies, crop yields, and scale of agricultural activity pursued by farms confirm the
favorable finding regarding the program's efficacy. These provide insight into the mechanisms through
which income increases were obtained. Income increases of PTT participant farms appear mainly due
to increases in the intensity of farming activity carried out by participants. This is shown by the
statistically significant effect of PTT participation on the area planted and the amount of labor applied
to farming among participant households. Evidence of program success in prompting participants to
move away from traditional cropping patterns and to adopt crops with more favorable economic
prospects and less deleterious effects on soil quality was less clear. Results concerning the PTT's
effectiveness in increasing crop yields and agricultural practice were also mixed.
Ahn, H. and Powell, J., 1993. Semiparametric estimation of censored selection models with a
non-parametric selection mechanism. Journal of Econometrics 58(1-2), 3-29.
Bebbington, A. and Sotomayor, O., 1998. Demand-led and poverty oriented…or just
subcontracted and efficient? Journal of International Development 10(1), 17-34.
Berdegué, J., 1994. Chile’s privatized extension system: 17 years of experience. Proceedings of
the International Symposium on Systems-Based Approaches to Agricultural and Rural
Development, Montpellier, France, 21-25 November, 1994.
Butler, J. and Moffit, R., 1982. A computationally efficient quadrature procedure for the one
factor multinomial probit model. Econometrica 50(3), 761-764.
Carney, D., 1998. Changing public and private roles in agricultural service provision. Overseas
Development Institute: London. 90 pp.
Chamberlain, G., 1984. Panel data. In. In Griliches, Z. and Intriligator, M. (Eds.). Handbook of
Econometrics, 2. North-Holland Publishers, Amsterdam, pp.1247-1318.
Edmonds, C., 1998. “Policy regimes, agrarian institutions, and the performance of smallholder
agriculture in Chile: Three essays analyzing longitudinal survey data on Chilean peasant farms
(1986 to 1995).” Ph.D. Thesis, University of California, Berkeley.
Falaha-Lumi, B., 1992. Una evaluación critica de los programas de transferencía tecnológica en el
agro Chileno. Documentos de Trabajo--Estudios Sociales, Facultad Latinoamericana de Sciencias
Sociales (FLACSO), Santiago, Chile.
Greene, W.H., 1990. Econometric analysis (2nd edition). Macmillan Publishing Company, New
York, 791 pp.
Greene, W.H., 1995. LIMDEP version 7.0: User's manual. Econometric Software Inc., Bellport,
Heckman, J., 1979. Sample selection bias as a specification error. Econometrica 46, 153-161.
Hsiao, C., 1986. Analysis of panel data, (Econometric Society Monographs No. 11). Cambridge
University Press, New York 246 pp.
Jakubson, G., 1991. Estimation and testing of the union wage effect using panel data. Review of
Economic Studies 58, 971-991.
Kennedy, P.E., 1981. Estimation with correctly interpreted dummy variables in semi-logarithmic
equations. American Economic Review 71(4), 801-802.
Lee, L.F., 1983. Generalized econometric models with selectivity. Econometrica 51, 507-512.
Leiva, C., and Sotomayor, O., 1994. Organzaciones campesinas, programas publicos y ONG: el
desafio de la articulación institucional. Grupo de Investigaciones Agrarias (GIA), Santiago, Chile.
Lopez, R., 1994. Rural poverty in Chile: A quantitative analysis. Manuscript, University of
Maryland, College Park, Maryland.
Maddala, G.S., 1986. Limited dependent and qualitative variables in economics. Cambridge
University Press, New York, 401 pp.
MIDEPLAN, 1991. Un proceso de integración al desarrollo: Informe Social 1990-1991. Ministry
of Planning (MIDEPLAN), Santiago, Chile.
MIDEPLAN, 1992. Poblacion, educacion, vivienda, salud, empleo, y pobreza: CASEN 1990.
Ministry of Planning (MIDEPLAN), Santiago, Chile.
Singh, I., Squire, L., and Strauss, J., 1986. Agricultural household models: Extensions,
applications, and policy. Johns Hopkins University Press, Baltimore, Maryland, 335 pp.
Sotomayor, O., 1994. Politicas de modernización y reconversión de la pequeña agricultura
tradicional Chilena. ODEPA and the Instituto Interamericano para la Agricultura (IICA),
World Bank, 1995. Chile, estrategia para elevar la competitividad agricóla y aliviar la pobreza
rural. World Bank Country Study, World Bank, Washington, DC.
The author wishes to thank Jeff Perloff, Irma Adelman, Ron Lee, and Alain de Janvry for their
considerable assistance during the many phases of this research. Thanks are also due to the many
Chilean researchers and agricultural extension practitioners without whose assistance this work
could not have been completed. Any errors or omissions are solely the author’s responsibility.