Off-farm labour decision of Canadian farm operators: Urbanization effects and rural labour market linkages

lable at ScienceDirect

Journal of Rural Studies 25 (2009) 12–24

Contents lists avai

Journal of Rural Studies

journal homepage: www.elsevier .com/locate/ j rurstud

Off-farm labour decision of Canadian farm operators: Urbanizationeffects and rural labour market linkages

Alessandro Alasia a, Alfons Weersink b,*, Ray D. Bollman a, John Cranfield b

a Agriculture Division, Statistics Canada, 170 Tunney’s Pasture Driveway, Ottawa, Ontario K1A 0T6, Canadab Department of Food, Agricultural and Resource Economics, University of Guelph, Gordon Street, Guelph, Ontario N1G 2W1, Canada

Keywords:Off-farm labourLabour marketsRural development

* Corresponding author. Tel.: þ1 519 824 4120.E-mail addresses: [email protected]

uoguelph.ca (A. Weersink), [email protected] (J. Cranfield).

0743-0167/$ – see front matter � 2008 Elsevier Ltd.doi:10.1016/j.jrurstud.2008.04.002

a b s t r a c t

Understanding the factors affecting off-farm labour decisions of census-farm operators has significantimplications for rural development and farm income support policy. We examine the off-farm labourdecisions of Canadian farm operators using micro-level data from the 2001 Census of Agriculturecombined with community level data from the 2001 Census of Population. While confirming some of thefindings of previous research with respect to the effects of human capital and farm characteristics on off-farm work participation, this study shows the differential impact of those variables for operators ofsmaller and larger holdings. Family, community and regional characteristics appear more relevant indetermining the joint decision to work off-farm and operate a smaller holding, compared to the decisionto work off-farm and operate a larger farm. Results suggest that, once other factors are accounted for,proximity to urban centres does not have a positive effect on the joint decisions to participate in off-farmwork and to operate a holding. This joint decision, in fact, is more related to the dynamics of the locallabour market. A major implication of these findings is that while urban centers might represent anengine of growth for overall rural income through employment opportunities for the non-farm work-force, the non-farm income of farm operators is more likely to be affected by policy initiatives thataddress directly the dynamics of labour markets in the community where the operator lives.

� 2008 Elsevier Ltd. All rights reserved.

1. Introduction

The total number of Canadian census-farm operators and thenumber of them reporting off-farm work are trending in oppositedirections. Between 1991 and 2001, the number of census-farmoperators fell by approximately 11%, from 390,870 to 346,195. Overthe same period, the number of operators reporting off-farm workrose approximately 6%, from 145,005 to 154,215 (Statistics Canada,2002). Similar indicators, such as the share of census-farm opera-tors working full-time off the farm and the contribution of off-farmincome to total family income for farm households, point to thelong-term structural shift in rural employment away from agri-culture and to the growing income diversification of families as-sociated with census-farms (Fuller and Bollman, 1992; Bollmanet al., 1992).

In the emerging rural economy, off-farm employment can arisefrom different motivations. First, engaging in off-farm employment

(A. Alasia), aweersin@(R.D. Bollman), jcranfie@

All rights reserved.

can be a self-insurance mechanism for households associated withan agricultural holding to help to stabilize total household incomegiven the inherent variability in net farm income. Second, off-farmemployment may be necessary to provide sufficient income tocover family living expenses if the operator of the farm is unable togenerate enough revenue to support a family. Third, off-farm labourmay be the primary household employment for some residents,who have chosen a rural lifestyle but, due to some agriculturalsales, they are recorded as operating a census-farm.

The factors affecting off-farm labour participation will likelyvary with farm size. Indeed, national statistics from the Census ofAgriculture show a striking difference in off-farm labour partici-pation by operation size (Statistics Canada, 2002). In 2001, about48% of operators of holdings with less than $250,000 in grossrevenue reported off-farm work while only 19% of operators thatgenerated $250,000 or more in gross farm receipts reported off-farm work. These participation numbers are consistent with thefindings of previous research that has included farm size as onefactor in the decision to work off-farm (Benjamin et al., 1996;Bollman, 1979; Findeis et al., 1991; Howard and Swidersky, 2000;Weersink, 1992). However, the existence of different motivationsbehind the choice of off-farm labour participation – particularly asan income level or stabilization objective versus potential rural

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–24 13

lifestyle choice – suggests the potential for a differential effect ofvarious mitigating and motivating factors based on farm size.Analysis of sub-samples delineated by farm size enables one todetermine if behaviour by small farm operators differs from that oflarge farm operators. The delineation allows for the testing of thehypothesis that human capital, farm, family and market charac-teristics have a larger relative effect on off-farm employment forsmaller operations and on the well-being of the family. Support forsuch a hypothesis would highlight the importance of pluriactivityas a strategy for maintaining household livelihoods (OECD, 1995).

In addition to examining the influence of factors such as locationon different types of farms, it is also instructive for policy to un-derstand the value of alternative labour markets. Studies such asGardner (2005) suggest that household income in rural areas isincreasingly determined by labour markets rather than the agri-cultural sector. Given the growth in employment opportunitiestends to be concentrated in urban regions, the link between urbanlabour markets and rural populations remain vital for the economicsustainability of rural areas (Bollman and Biggs, 1992; Schindeggerand Krajasits, 1997). However, an alternative hypothesis is that theurban influence on rural well-being is declining and that a greaterrural-to-rural pattern of interaction is emerging (Green and Meyer,1997). While healthy labour markets are a key means of improvingthe financial well being of farm households, the focus of the policyinitiatives to enhance these markets depends if farm operators areinfluenced by urban or rural labour markets.

In this paper, we examine the off-farm labour decisions of Ca-nadian census-farm operators using micro-level data from the 2001Census of Agriculture combined with community-level data fromthe 2001 Census of Population. We use a set of spatially laggedcovariates which allow us to assess the effect of regional charac-teristics on off-farm labour decisions and estimate the model for‘‘smaller’’ and ‘‘larger’’ census-farms, separately. After presentingthe empirical model, the description of the variables, and theirhypothesized effects in the following section, we elaborate furtheron the findings, particularly the difference in importance of labourmarkets in rural versus urban areas on off-farm labour decisions,and their implications on policy. One important contribution to theliterature is that family, community and regional characteristics aremore relevant in determining the joint decision to work off-farmand operate a smaller holding, compared to the decision to workoff-farm and operate a larger farm. Second, once other factors areaccounted for, proximity to urban centres does not have a positiveeffect on off-farm employment. Instead, it is rural labour marketsthat are important for farm operators involved in off-farm work,particularly smaller ones. The results suggest that if family farms,typically smaller holdings, remain the back bone of agricultural andrural development policies, then policy initiatives should paygreater attention to the dynamics of rural labour markets.

2. Model specification and estimation

The off-farm labour decision by a farm operator is typicallydescribed though a household production model in which the in-dividual maximizes utility derived from purchased goods and fromleisure through the allocation of labour (Bollman, 1979; Huffman,1991; Weersink, 1992). The optimal amount of farm work is wherethe incremental value of extra time on the farm is equal to themarginal rate of substitution between leisure and consumption. Ifthe wage rate is greater than the marginal returns to farm work,then the farmer will engage in off-farm employment until the in-cremental returns to both forms of employment are equated. Sincethe work allocation between farm and off-farm is determinedjointly, the decision to participate in off-farm employment isa function of all exogenous variables in the household productionmodel: human capital, family, farm, and market characteristics.

2.1. Econometric model

Model specification requires the identification of an appropriatefunctional form for estimation and the definition of operationalvariables that reflect the underlying theoretical concepts. Ideally,the opportunity cost of off-farm labour would be estimated asa function of the explanatory variables suggested by theory. How-ever, the reservation wage that determines off-farm labour partic-ipation cannot be measured. Moreover, available data from theCensus of Agriculture do not allow measurement of off-farm labourallocation as a continuous variable. Typically, off-farm labour par-ticipation is recorded as a categorical outcome, either participationor non-participation in off-farm work.

The characteristics of this problem and the focus of this studyare suited for qualitative dependent variable modeling (Long andFreese, 2001). Following this approach, the reservation wage can beassumed to represent a latent model, which is not observable inreality, but which can be related to an observable model in whichthe dependent variable takes a dichotomous outcome. In this case,what is observed is whether the operator reports off-farm work ornot. We use a probit specification to model the probability of thisevent. The probability of an observation (census-farm operator) tobelong to one of two mutually exclusive groups (off-farm workversus no off-farm work) is defined by the integration of the normalprobability density function, N(0,1), between �N and a linearfunction of the independent variables (xb). The specification of themodel is as follows,

Pr�y ¼ 1 xj Þ ¼

Z xb

�N

1ffiffiffiffiffiffi2pp exp �u2

2

� �du ¼ F xbð Þ (1)

where y is the vector of independent binary outcomes assumed tobe mutually exclusive and exhaustive, x is the set of explanatoryvariables which includes individual, family, characteristics of theagricultural holding and community level variables, and F(.) de-notes the cumulative normal distribution function (cdf).

2.2. Effect of covariate changes on the probability of off-farm labour

Probit models imply a non-linear relationship between the ex-planatory variables and the dichotomous dependent variable,which is defined by the cumulative distribution function F(.). Sincethe model estimates the conditional probability Pr(yjx) of off-farmlabour participation, non-linearity is in fact a desirable property.The challenges posed by this specification, however, are several.Besides the fact that the coefficients cannot be directly interpretedwith any substantive meaning, the magnitude of change in thepredicted probability associated with a given change in one of theexplanatory variables depends on the values of all explanatoryvariables, and must be evaluated at given values of x. There is nosingle approach that can be used to summarize the complexity ofthese relationships (Long and Freese, 2001).

Conventionally, the marginal effects are used to display the re-lationship between single explanatory variables and the probabilityof a certain outcome. Unlike the standard regression which hasconstant marginal effects equal to the regression coefficients, themarginal effects in a probit model are non-linear and their valuesdepend on the level of all explanatory variables, x, at which they areevaluated. Consequently, the marginal change can be substantiallydifferent from the probability change that would result from a smalldiscrete variation of xk. For this reason, we paygreater attention to theeffect of discrete changes in the independent variables. We computethe probability associated with specific values of the explanatoryvariables and then calculate the difference between the predictedprobabilities of these ‘‘typical cases’’. Thus, for a discrete change of xk

the associated probability change is given by the following,

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–2414

DPr y¼ 1ð jxÞDxk

¼ Pr y¼ 1ð jx; xk þ dÞ � Pr y¼ 1ð jx; xkÞ (2)

where d represents the discrete change of xk and the rest is aspreviously defined. The computation is done by keeping all otherexplanatory variables at their sample means.1

Using this approach, we evaluate the probability of off-farm la-bour at minimum (xmin) and maximum (xmax) values for each ex-planatory variable, and the corresponding probability changebetween these two values. For continuous variables, we also com-pute the probability change associated with one standard deviation(SD) change of the explanatory variable around its mean (m). Hence,we evaluate the probability at values of x¼ m � (a $ SD) andx¼ mþ (l $ SD), where a, l> 0 and aþ l¼ 1. For the variables thatare approximately normally distributed, we set the value of a and l

equal to 0.5, hence x¼ m� (1/2 $ SD). For the variables that havea skewed distribution, we use the value of a¼ 0.1 and l¼ 0.9, to avoidcomputing the probability associated with a negative value of x.2

2.3. Specification of the spatially lagged variables

A key explanatory variable is location and we distinguish twocomponents of the regional milieu effect: the local effect and theregional effect. This distinction is introduced by means of CensusConsolidated Subdivision (CCS) variables, to capture the effect oflocal characteristics, and their corresponding spatial lag, to capturethe regional effect. For each CCS and indicator of interest, the spatiallag is a distance-weighted average of the neighboring CCSs’ valuesfor that given indicator.

In general terms, the computation of spatial lags involves threesteps. First, a spatial weight matrix W is computed using proximitycriteria between the geographic units of analysis (CCSs in our case).The elements wi,j of the squared matrix W defines the nature of thespatial relationships among the geographical units; hence, its spec-ification is of particular importance. There are alternative ways inwhich the matrix can be specified and relatively limited theoreticalguidance is available for choosing among them (Anselin, 1988).Typically, the choice rotates around alternative definitions ofboundary proximity or geographic distance and it is common prac-tice to assess the robustness of the analysis by testing the effects ofalternative specifications of the weight matrix (see for instanceBoarnet et al., 2005). Second, the spatial weight matrix is row stan-dardized; the elements of each row are scaled so that the rows sum toone, resulting in a form of spatial smoothing. Finally, the W matrix isused to compute spatially lagged variables (x_lag), by multiplying thenxn matrix, W, with the nx1 vector of the indicator of interest.

For the problem at hand, a distance measure between territorialunits appears more appropriate than any boundary criteria to definespatial relationships. Hence, the three steps followed in the com-putation of the spatial lag variables can be summarized as follows,

wi;j ¼ 1=dai;j isj di;j � D

wi;j ¼ 0 otherwise

(

wsi;j ¼ wi;j=

Pj

wi;j

x lagi ¼X

j

wsi;j,xj (3)

The element wi,j of the spatial weight matrix is set equal to theinverse of the squared distance (d) between each pair of CCS’s

1 When the explanatory variable is introduced in both linear and quadratic form,such as the case of age and age squared, both terms are included in the computationof the variable’s effect.

2 The variables for which we use a¼ 0.1 and l¼ 0.9 are the following: Area,Capital, Sales, and Hired work.

geographic centroids (i s j), while for i¼ j the element is equal tozero. The attrition parameter, a, is used to amplify or reduce theeffect of the distance and in our computation this is set equal to 2,while the bandwidth for spatial interaction (D) was set equal to1000 km.3

3. Data

3.1. Variable selection

The model in Eq. (1) is estimated using micro-data combinedwith community data at the CCS level. Micro-level data are from theCensus of Agriculture-Population Linkage database 2001, whichlinks information for a 20% sample of operators, as identified in theCensus of Agriculture, with their socio-economic information col-lected from the Census of Population. This sample includes 70,851census-farms operators. Community indicators and their corre-sponding spatial lags are based on data from the 2001 Census ofPopulation. After excluding the operators in collective dwellings(513 observations) and combining the two datasets, the resultingdataset for estimation encompassed 69,797 farm operators and1746 CCSs. The model was estimated for two sub-samples, corre-sponding to 58,212 operators of smaller census-farms, defined asoperators of holdings with gross receipts less than $250,000, and11,585 operators of larger census farms, defined as operators ofholdings with gross receipts equal to or greater than $250,000.

For individuals who operate a census-farm, the decision toparticipate in off-farm employment is assumed to be affected byindividual, family, farm, local and regional labour market charac-teristics. Although some degree of endogeneity is difficult to avoidin this type of analysis, the variables were selected so as to mini-mize the problem (for instance, household income was not in-cluded in the model). A technical description of the model variablesis provided in Table 1. Below we provide a concise description ofthese variables, along with the rationale for their use and theirexpected effect.

Human capital theory suggests that the marginal productivity ofany self-employed operator will be affected by socio-economicfactors such as age, gender, and education (Vera-Toscano et al.,2003, 2004). Although both the off-farm wage rate that can beearned by the operator and the incremental value of farm workshould increase with human capital, previous studies have gener-ally found the relative effect larger for off-farm employment. Age ofthe operator is included as a quadratic variable since it is assumedthat more experience increases the relative employability of theoperator but there comes a point when the individual may not beable to handle the tasks or to adjust to new skills as well asa younger person. Gender of the farm operator is included to reflectpotential discrimination against females, which may be moreprevalent in rural labor markets (Phimister et al., 2002). Educationis measured by four dummy variables representing the highestlevel of schooling attained by the operator with the base variablebeing less than high school. Higher education levels have beenassociated with higher wage levels in rural Canada (Vera-Toscanoet al., 2003) so these higher off-farm returns will increase thelikelihood of off-farm employment for operators with higher levelsof educational attainment.

Family characteristics include the presence of young children,mobility status, and the amount of time spent in domestic work.

3 It should be noted that for other typical value of the attrition factor a (that isa¼ 1 and 3) the results of the analysis were virtually identical. The choice of thisbandwidth (D) was dictated primarily by a computational consideration; this valueis slightly above the minimum distance that allows each observation to interactwith at least one regional neighbor.

Table 1Description and computation of variable used in the model

Dimension/variable Detailed description and computation

Off-farm work (dependent) Dichotomous variable which takes value of 1 if the operator report any work off the census-farm holding, regardless of the amount of time spentin off-farm work. In the Census of Agriculture 2001, ‘‘operators’’ were defined as those persons responsible for the day-to-day managementdecisions made in the operation of a census-farm or agricultural operation. Up to three operators could be reported per census-farm.

Individual characteristicsAge The age of the operator is entered as number of years (Age), as well as in its quadratic form as squared number of years of the operator (Age ) Age)

to allow for diminishing, or increasing, effects of an additional year of age.Gender Gender of the farm operator is entered as dummy which takes value of 1 if the operator is female, and 0 otherwise.Education A set of four dummy variables account for the effect of educational attainments: Less than High School Diploma takes the value 1 if the operator

has not completed high school, and 0 otherwise (omitted category); High School takes the value 1 if the operator has a secondary schoolcertificate or non-university degree; University certificates takes value of 1 if some university education below bachelor level, with or withoutdiploma, is reported; University degree takes value 1 if the operator has a bachelor degree or higher university degrees, and 0 otherwise.

Family characteristicsChildren Dummy variable that takes value of 1 if the presence of one or more children less than 15 year of age is reported in the family of the operator, and

0 otherwise.Mobility Dummy variable that takes value 1 if the operator has changed address (dwelling) over the past five years, and takes value 0 is the operator has

not changed address.Domestic work Three independent dummy variables capture the extent to which the operator is involved in domestic works. Unpaid housework takes value 1 if

the operator does more than 15 h/week of unpaid housework, and value 0 otherwise; Unpaid childcare takes value of 1 if the operator is involvedin more than 5 h/week of child caring, and 0 otherwise; Unpaid senior care takes value 1 if the operator is involved in more than 5 h a week of careto seniors.

Farm characteristicsFarm type A set of twelve dummy variables describe the type of farming operation. This classification is generated by the Census of Agriculture (Statistics

Canada, 2002). Each census-farm is classified according to the predominant type of production. This is done by estimating the potential receiptsfrom the inventories of crops and livestock reported on the census questionnaire and determining the product or group of products that make upthe majority of the estimated receipts. The resulting twelve groups are the following (the dummy takes value 1 if the farm belong to the groupand 0 otherwise): Dairy (omitted category), Cattle (Beef), Hog, Poultry and egg, Wheat, Grain and oilseed (except wheat specialty farms), Field crop(except grain and oilseed specialty farms), Fruit, Miscellaneous specialty, Livestock combination, Vegetable, Other combination.

Number of farm operators A set of three dummy variables indicates the number of operators involved in the census-farm (1 if the operators belong to that group, and0 otherwise); the dummies are: Single operator (omitted category), Two operators, Three operators.

Business structure Three types of farm organization are distinguished by using of a set of dummy variables (1 if the operators belong to that group, and 0 otherwise):Sole proprietorship (omitted category); Partnership (with or without written agreement); Corporation (including family and non-family).

Farm size Three indicators of farms size are used both in the linear and quadratic form to allow for diminishing, or increasing, effects of farm size. Area is thetotal land area of the farm in thousand hectares; Capital is the total farm capital in million dollars; Sales is the total gross farm receipts (excludingforest products) in thousand dollars.

Hired workers This variable, in linear and quadratic form, is the total number of weeks of hired work, from both part-time and seasonal workers, used on thecensus-farm during the census year.

Local and regional characteristicsEmployment change The local employment change is computed at the CCS level as percentage change of total experienced labour force between 1991 and 2001.

Regional employment change is the corresponding spatial lag.Specialization The local specialization of the economy is measured by the Herfindahl index applied to experienced labour force data; the index is the sum of the

squares of the industry employment shares in 2001 (experienced labour force). Eleven major industry groups are used in the computation, whichinclude: agriculture, other primary sectors, four type of manufacturing industries (natural resource related (NRR), labour intensive (LI), scalebased (SB), and product differentiated and science based (PDSB)), construction, and four type of service industries (distributive, business,consumer, and public services). Regional specialization is the corresponding spatial lag.

Unemployment rate The local unemployment rate of each CCs is computed as total unemployed individuals age 25–54 divided by unemployed and employedindividuals age 25–54; Regional unemployment rate is the corresponding spatial lag.

Employment shares Following Findeis et al. (1991), we focus on manufacturing and service industries. Eight industries types are used, four for manufacturing andfour for services. Industries are classified according to NAICS 2001 system. We use a classification of industries based on statistics Canada andprevious research of the authors (details are available form the authors upon request). Manufacturing is divided into Natural Resource Related(NRR), Labour Intensive (LI), Scale Based (SB), and Product Differentiated and Science Based (PDSB); service industries are divided into distributive,business, consumer and public services. Each indicator is used at the local and regional level (spatial lag).

Distance to CA/CMA This is the distance in km between CCS geographic centroid and the geographic centroid of the nearest Census Agglomeration (CA) or CensusMetropolitan Agglomeration (CMA). The variable is used both in the linear and quadratic form.

Population density Local population density is computed as total non-institutional population of a CCS in 2001 divided total land area of the CCS; Regional populationdensity is the corresponding spatial lag.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–24 15

The reservation wage of the operator is assumed to increase withchildren in the family since the value of home time is increasedrelative to the time spent working. However, having young childrenmay increase the financial needs of the family and thus may promptthe necessity for additional income from farm or off-farm work. Thecosts of searching for employment and the adjustment costs for re-location are assumed to decrease, and thus, the likelihood of off-farm employment to increase, if the family has moved previously.Engagement in domestic work may also affect off-farm labourdecisions. Even though there is limited theoretical indication on thepotential effect of these variables, we include engagement inhousework, and child and senior care as control variables.

Characteristics of the census-farm holdings include the type offarming operation. Excess labour available for off-farm employ-ment will decrease with the labour requirements on the farm andthese requirements are more likely to be higher for livestock op-erations such as dairy and hog farms that require daily efforts thanfor crop farms where labour needs are largely seasonal. The jointoperation of a census-farm holding and participation in off-farmwork is expected to be largest for holdings that are easiest tooperate on a part-time basis – such as beef cattle or horses (whichmay, in fact, be pets rather than animals for sale).

Increasing the number of operators in the farm is assumed toincrease the likelihood of off-farm employment since the constraint

4 In the rest of the paper we will use the term ‘‘average farm operator’’, or simplyaverage operator, to indicate the results for the full sample model with averagevalues for all the explanatory variables.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–2416

on labour availability is relaxed. Similarly, alternative businessstructures may be more conducive to working off the farm. Apartnership or corporation may increase farm productivity throughlabour specialization and thus decrease the likelihood of off-farmemployment (Rosen, 1978). Similarly, the specialization may de-lineate the roles of investor and operator in which case the lattermay be less likely to engage in off-farm employment.

Increases in farm size as measured by land area or asset valueare assumed to increase the marginal productivity of farm labourand subsequently lower the probability of working off the farm. Theincremental value of farm labour may also be increased throughhired workers that allow for specialization of the operator in moremanagement-related activities. While hired labour could bea complement to the labour time spent on the farm by the operatoras supported by Benjamin et al. (1996) and Bollman (1979), it couldalso serve as a substitute and allow the operator to work for highermarket wages than those paid to the farm employees.

Local and regional characteristics reflect the strength of the jobmarket and institutional factors affecting both the demand andsupply of non-farm employment. We distinguish between local andregional effects by using community level variables and their cor-responding spatially lagged variables, as discussed in the previoussection. For instance, the unemployment rate of the community inwhich the farm operation is located proxies the strength of thelabour market at the local level. Although the local dimensionmight be critical in determining employment outcomes of farmoperators, an alternative hypothesis is that the conditions of theregional labour market have a greater bearing in affecting off-farmemployment outcome. The inclusion of local and spatially lagged(regional) variables allows us to explore this dimension. Besidesunemployment, total employment growth and prevalence ofmanufacturing and service industries have been used to assess thestrength of labour markets (Findeis et al., 1991). To capture thisaffect, we use total employment change, in percentage terms, at thelocal and regional level between 1991 and 2001. Labour marketdemand characteristics are proxied by the employment shares foreight industry groups at the local and regional level. If a region ishome to primarily low-paying service jobs, then the relative valueof the market wage opportunities will decrease relative to the valueof farm work and thus, so will the likelihood of off-farm employ-ment. The Herfindahl index of concentration was also used tocapture the degree of employment specialization of a locality andregion. It is expected that less diversified economies offer less op-portunities for off-farm employment.

Finally, the size of the labour market and its density can affectthe likelihood of off-farm employment. We include two measuresof urbanization to assess these potential effects: the distance to anurban centre, in its linear and quadratic form, and populationdensity at the local and regional level. A common expectation isthat agglomeration economies associated with reduced transportcosts to work (i.e. transportation cost of people) and a greater va-riety of employment opportunities increase the relative value ofoff-farm employment in more urbanized areas (Glaeser and Mare,2001), while rural areas are more likely to have higher search costsfor employment, more occupational mismatches, and fewer jobtraining opportunities (Wheeler, 2001; Findeis and Jensen, 1998).The existing literature, however, does not appear to provide a clearindication on the effect of urbanization factors on off-farm labourdecisions (Findeis et al., 1991). As we discuss in the next section, ourfindings provide an alternative perspective on the effects ofurbanization.

4. Results

Summary statistics for the variable explaining off farm em-ployment are given in Tables 2 (for the full sample), 3 and 4 (for the

sub-samples). The level and change in predicted probabilities as-sociated with selected values of the explanatory variables, for thefull sample and for operators of smaller and larger census-farms aregiven in Tables 4–6, respectively. The figures in bold correspond toregression coefficients that are statistically significant at the 90%confidence level or higher. The discussion of the results focuses onthese tables, with specific attention to the differences betweenoperators of smaller and larger farms. The coefficients of the probitmodels are not reported due to limited space and interpretability,but are available from the authors by request. It should be men-tioned, however, that the measures of fit of the models are rea-sonably good for cross-sectional data (McFadden‘s R2 value is 0.19,0.23, 0.07 for the full sample, operators of smaller and larger censusfarms, respectively; McKelvey and Zavoina’s R2 value is 0.38, 0.42,0.14 for the same three groups, respectively).

Three major results emerge from the alternative estimates. First,human capital characteristics, and specifically high educationalattainment, are always associated with greater likelihood of beingemployed off-farm. Second, for both the operators of smaller andlarger holdings, there is an inverse and significant relationshipbetween farm size and off-farm employment (although farmsize is clearly less relevant for the sub-sample of larger farms).Third, family and locational characteristics, including urbanizationfactors, appear more important in determining the off-farmdecisions for operators of smaller farms than those of larger units.

Age has an important effect on the probability of being engagedin off-farm work, particularly for operators of smaller farms.Compared to the average farm operator4 who has a predictedprobability of off-farm work of 0.41, the youngest operators areabout 2% more likely to be engaged in off-farm work. The re-lationship between age and off-farm work, however, is not linear.Other conditions held at the average, this probability peaks at ap-proximately 35 years of age (Pr¼ 0.59) and drops rapidly for themost senior operators in the sample. For the average operator,a change in age of 13 years (one standard deviation) from about 43to 56 years of age is associated with a probability decline of 18%.Similar effects are noted for operators of smaller farms, while age isnot a significant variable for larger operators.

Gender also has a significant effect on the likelihood of off-farmemployment and as with age, the relative effect varies with farmsize. Compared to the average male operator, the average femaleoperator is 6% less likely to be engaged in off-farm work. However,average female operators of smaller farms are about 10% less likelyto work off-farm compared to their male counterparts, while fe-male operators of larger units are almost 7% more likely to work off-farm compared to male operators of similar size units.

The probability of operating a census-farm and working off-farm is particularly high if the operator has a university degree.Compared to the average operator, the average farmer with a uni-versity degree is almost 20% more likely to work off-farm(Pr¼ 0.60). For operators of larger farms, this probability differen-tial reduces to about 9%. The positive effect of educational attain-ment increases with the level of schooling and is consistent withprevious studies that find the increases in marginal returns fromeducation are higher for off-farm employment than farm work.

Family characteristics only have significant effects for operatorsof smaller farms (and for the full sample model, which is largelyaffected by smaller farms since these farms constitute 83% of thefull sample). Compared to the average operator, individuals whohave recently moved into the community are about 4% more likelyto be involved in off-farm work. The move may have been

Table 2Sample descriptive statistics: full sample

Variable No off-farm work Off-farm work Sample

Mean (or ratio) SD Mean (or ratio) SD Mean (or ratio) SD

Individual characteristicsAge 53.1 13.975 45.8 10.827 49.9 13.186Gender 0.259 0.438 0.276 0.447 0.267 0.442Education Less than high school 0.445 0.497 0.272 0.445 0.369 0.482

High school 0.421 0.494 0.520 0.500 0.465 0.499University certificates 0.070 0.255 0.089 0.285 0.078 0.269University degree 0.063 0.244 0.118 0.323 0.088 0.283

Family characteristicsPresence of children 0.267 0.442 0.392 0.488 0.322 0.467Mobility 0.119 0.324 0.195 0.396 0.153 0.360Domestic work Unpaid housework 0.435 0.496 0.454 0.498 0.443 0.497

Unpaid childcare 0.272 0.445 0.374 0.484 0.317 0.465Unpaid senior care 0.080 0.271 0.082 0.274 0.081 0.272

Farm characteristicsNumber of farm operators One operator 0.447 0.497 0.433 0.495 0.441 0.496

Two operators 0.451 0.498 0.506 0.500 0.475 0.499Three operators 0.102 0.302 0.061 0.239 0.084 0.277

Farm type Dairy 0.142 0.349 0.024 0.153 0.090 0.286Cattle (beef) 0.265 0.441 0.315 0.464 0.287 0.452Hog 0.043 0.203 0.022 0.147 0.034 0.180Poultry and egg 0.023 0.149 0.021 0.144 0.022 0.147Wheat 0.057 0.232 0.052 0.221 0.055 0.227Grain and oilseed (no wheat) 0.203 0.402 0.195 0.396 0.200 0.400Field crop (no grain & oilseed) 0.075 0.263 0.094 0.291 0.083 0.276Fruit 0.028 0.166 0.034 0.180 0.031 0.173Miscellaneous specialty 0.104 0.306 0.183 0.387 0.139 0.346Livestock combination 0.018 0.134 0.026 0.159 0.022 0.146Vegetable 0.015 0.123 0.012 0.110 0.014 0.117Other combination 0.026 0.159 0.022 0.148 0.024 0.154

Business structure Sole proprietorship 0.446 0.497 0.514 0.500 0.476 0.499Partnership 0.347 0.476 0.380 0.485 0.362 0.481Corporation 0.207 0.405 0.105 0.307 0.162 0.368

Farm size Total land area (1000 ha) 0.411 1.269 0.219 0.629 0.326 1.040Total capital ($1,000,000) 1.195 2.935 0.634 1.477 0.947 2.417Total sales ($1000) 339.227 2635.817 100.273 807.471 233.367 2042.687

Hired workers Total weeks of hired work 64.051 384.565 18.086 176.637 43.7 311.0Local and regional characteristics

Local employment growth 8.035 15.969 9.276 16.067 8.585 16.024Regional employment growth 6.663 8.859 7.079 8.955 6.847 8.904Local specialization 0.188 0.071 0.185 0.062 0.186 0.067Regional specialization 0.199 0.045 0.200 0.042 0.200 0.044Local unemployment 0.052 0.043 0.053 0.043 0.053 0.043Regional unemployment 0.054 0.032 0.054 0.031 0.054 0.031

Employment structure Local manufacturing NRR 0.039 0.035 0.036 0.031 0.037 0.033Regional manufacturing NRR 0.037 0.021 0.034 0.019 0.035 0.020Local manufacturing LI 0.022 0.033 0.018 0.029 0.020 0.031Regional manufacturing LI 0.022 0.025 0.018 0.020 0.020 0.023Local manufacturing SB 0.038 0.040 0.039 0.040 0.038 0.040Regional manufacturing SB 0.038 0.028 0.036 0.027 0.037 0.028Local manufacturing PDSB 0.023 0.024 0.021 0.022 0.022 0.023Regional manufacturing PDSB 0.023 0.016 0.021 0.015 0.022 0.015Local distributive services 0.110 0.036 0.111 0.034 0.110 0.035Regional distributive services 0.109 0.015 0.109 0.014 0.109 0.015Local business services 0.069 0.035 0.071 0.034 0.070 0.035Regional business services 0.071 0.022 0.072 0.022 0.071 0.022Local consumer services 0.202 0.057 0.208 0.056 0.205 0.057Regional consumer services 0.204 0.037 0.209 0.038 0.206 0.037Local public services 0.207 0.066 0.214 0.064 0.210 0.065Regional public services 0.208 0.030 0.212 0.030 0.210 0.030

Urbanization Distance to CA/CMA 53.4 45.0 58.2 50.6 55.6 47.6Local population density 48.1 164.5 43.7 147.7 46.2 157.3Regional population density 99.1 164.7 100.2 161.5 99.6 163.3

Number of operators in the sample 38,876 30,921 69,797

Note: these data refer to operators and operator families in private dwellings (residents of collective dwellings are excluded). The data are unweighted to be consistent with theprocedure adopted in the econometric model. The averages for local and regional characteristics should be interpreted as the average local and regional condition experiencedby the operators in the sample. Source: authors’ computations based on Census of Agriculture 2001 and Census of Population 2001 data.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–24 17

associated with the need for off-farm employment but it might alsoindicate lower adjustment costs relative to an individual who hasremained in one location. For average operators of smaller farmswho report both unpaid housework and senior care, the probability

of working off-farm is about 7% less than a similar operator notreporting any unpaid domestic labour.

Combining human capital and family determinants, a femaleoperator of a smaller farm reporting domestic work and less than

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–2418

a high school education has a less than 50% chance to be engaged inoff farm work, regardless of the age cohort.5 This may be due toeither additional household responsibilities that limit the amountof time available for market employment or to gender discrimina-tion by potential employers. In contrast, a male operator of a smallfarm with a university degree and no sizable engagement in do-mestic work has a greater than 70% probability of off-farm em-ployment for any cohort between 20 and 50 years of age. Thegender differentials tend to decrease for higher working age co-horts if the educational attainment is lower, but are more visiblypersistent across working age cohorts for higher educationalattainments.

The effect of farm characteristics on the probability of off-farmwork is remarkably consistent across samples, although the mag-nitudes of changes differ somewhat. The average operator ofa census-farm with more than one operator is 3–4% more likely towork off farm as compared to his/her counterpart who is a singleoperator on the farm. The figure increases to 6–8% for operators ofsmaller farms. Business arrangements also affect the likelihood ofoff-farm work; both average operators of smaller and larger in-corporated farms are about 3% less likely to engage in off-farmactivities, as compared to their counterparts involved in a partner-ship or sole proprietorship.

Some of the largest changes in the probability of off-farm labourare observed for operators of different farm types. The probabilityof off-farm engagement for the average small farmer not operatinga dairy farm increases between 23% for vegetable farm operators to37% for grain and oilseeds farm operators, as compared to a similaroperator who is dairy farming. For large farms, this probabilitychange is between 10% (vegetable farm) and 25% (poultry and egg).The effect of being a dairy operator on the probability of off-farmwork is striking; dairy operators are universally less likely to seekoff-farm employment.

The effect of discrete changes in farm size is also revealing.Changes between the minimum and maximum values of farm sizegenerally have a large impact on the probability of working off thefarm, but the effect on the probability is generally more limited forone standard deviation changes. Since total sales were used asa classification criterion to split the sample into small and largeholdings, it is not surprising that the magnitude of probabilitychange is different between estimates, with the effect of these in-dicators being generally more important for operators of smallerfarms. For the average operator of a small farm, total sales area critical factor in determining off-farm labour participation. Theaverage operator of a small farm with gross sales of $59,000 hasa 47% chance of working off-farm, while a comparable operatorwith gross sales of about $250,000 has a 26% chance. For both smalland large operators, a one standard deviation change in capital sizearound the mean (see Table 3 for descriptive statistics) decreasesthe likelihood of off-farm work by approximately 2%. A similarchange in the use of hired labour results in an approximate 1%decline in the probability of off-farm work (Tables 5 and 6) sug-gesting hired labour is a complement to operator labour rather thana substitute. The quadratic terms of the farm size indicators havea sizable effect only for extreme values of farm size (and for rangesin which confidence intervals tend to be very large).

Some of the most interesting findings of this analysis concernthe effects of location. Local and regional labour market charac-teristics are more relevant for operators of smaller farms. Themarginal effects for these employment variables are often large butthe effect of discrete changes are more limited, which confirmsthat, for continuous variables, marginal effects are not always the

5 Detail tabulations are available from the authors upon request.

best indicator of probability changes from changes in the cova-riates.6 Moreover, the analysis of probability changes for ‘‘typicalcases’’ that combine specific values of the explanatory variablesadds considerable information to the analysis of single coefficients.

Operators who live in communities that experienced rapidemployment growth are more likely to engage in off-farm work.The average operator of a smaller farm who lives in a communitywith the lowest employment dynamics is 6% less likely to work off-farm compared to a similar operator living in a community with thehighest employment growth over the last decade. The probabilitydifferential between a locality with an 8% employment decline andone with a 24% employment gain (one standard deviation change)is less than 1% (Table 5). The effect of changing regional economicspecialization and the regional unemployment rate is larger foroperators of smaller farms: about a 2% change for a standard de-viation change around the variable means.

When labour market indicators are combined to representtypical cases, their effect on off-farm engagement becomes evident.For the average operator that is located in a community with a weaklabour market (here defined by an employment decline of 5% overthe previous ten years, a regional unemployment rate of 20%, anda Herfindahl index of 0.2), the probability of off-farm employmentis below 40%. In contrast, the same average operator located ina community with strong labour market conditions (employmentgrowth of 50%, a regional unemployment rate of 6.6%, and a Her-findahl index of 0.15) has a slightly more than 50% chance ofworking off-farm.7

Exploring the alternative combinations of local/regionalmanufacturing and service employment indicates that both typesof economic activity have a positive effect on the probability ofoff-farm work (once the effects of their sub-sectors are combined).However, an increasing regional employment share in manu-facturing offsets the effect of local employment in manufacturing,while this is not the case for service employment. Once the effectof the manufacturing variable is combined, it appears that theaverage operator of a smaller farm, who lives in a community with30% of the labour force in manufacturing, has a probability of 0.46of working off the farm. This probability drops to 0.38 if the regionhas only 20% of employment in manufacturing.

The effect of urbanization indicators is particularly interestingand suggests that, once other factors are controlled for, off-farmemployment options for census-farm operators are not necessarilydetermined by urban labour markets. On the contrary, these jobopportunities appear to be more closely integrated with rural labourmarkets. For the average operator located in the closest proximity toan urban center, the probability of working off the farm is 39%(about 2% less than that of the average operator, located 55 km awayfrom an urban center). Moving from 25 km to 75 km away from anurban center (one standard deviation change), the predicted prob-ability for the average operator increases another 2% points, and fora similar operator located in the most remote areas (about 350 kmfrom an urban center) the chance of working off the farm are 9%higher than for the average Canadian operator. A similar pattern isobserved for the average operator of smaller farms. The likelihood ofworking off-farm for average operators located close to a CA/CMA(60 km) in a community and region with a high population density(3000 and 1600 inhabitant/km2, respectively) is about 0.3. Thiscontrasts with a predicted probability of almost 0.5 for a similaroperator located 320 km from a CA/CMA in a low density commu-nity and region (720 and 550 inhabitant/km2, respectively).

These results are counterintuitive but have plausible explana-tions. The common expectation is that individuals associated with

6 Detail tabulations are available from the authors upon request.7 Detail tabulations are available from the authors upon request.

Table 3Sample descriptive statistics: operators of smaller and larger census-farms

Variable Operators of smaller census-farms (sales <$250,000) Operators of larger census-farms (sales �$250,000)

No off-farm work Off-farm work No off-farm work Off-farm work

Mean (or ratio) SD Mean (or ratio) SD Mean (or ratio) SD Mean (or ratio) SD

Individual characteristicsAge 54.975 14.173 45.917 10.836 47.187 11.472 44.259 10.593Gender 0.274 0.446 0.274 0.446 0.211 0.408 0.299 0.458Less than high school 0.482 0.500 0.273 0.446 0.330 0.470 0.262 0.440High school 0.396 0.489 0.521 0.500 0.503 0.500 0.508 0.500University certificates 0.064 0.246 0.088 0.283 0.087 0.283 0.107 0.309University degree 0.058 0.234 0.118 0.322 0.080 0.271 0.122 0.328

Family characteristicsPresence of children 0.227 0.419 0.386 0.487 0.393 0.488 0.459 0.498Mobility 0.110 0.313 0.198 0.398 0.147 0.354 0.162 0.369Unpaid housework 0.474 0.499 0.461 0.498 0.311 0.463 0.371 0.483Unpaid childcare 0.251 0.433 0.372 0.483 0.339 0.473 0.401 0.490Unpaid senior care 0.086 0.280 0.083 0.276 0.062 0.240 0.065 0.246

Farm characteristicsOne operator 0.487 0.500 0.445 0.497 0.323 0.468 0.281 0.450Two operators 0.449 0.497 0.505 0.500 0.458 0.498 0.524 0.500Three operators 0.065 0.246 0.051 0.219 0.219 0.414 0.195 0.397Dairy 0.107 0.309 0.018 0.133 0.254 0.435 0.102 0.302Cattle (beef) 0.303 0.460 0.324 0.468 0.144 0.351 0.192 0.394Hog 0.022 0.145 0.016 0.127 0.110 0.313 0.095 0.294Poultry and egg 0.009 0.094 0.015 0.122 0.066 0.249 0.100 0.300Wheat 0.067 0.250 0.053 0.224 0.025 0.156 0.036 0.186Grain and oilseed (exc. wheat) 0.200 0.400 0.186 0.389 0.212 0.409 0.318 0.466Field crop (exc. grain & oilseed) 0.083 0.275 0.098 0.297 0.050 0.217 0.041 0.199Fruit 0.032 0.176 0.035 0.184 0.017 0.129 0.017 0.127Miscellaneous specialty 0.116 0.320 0.193 0.395 0.068 0.252 0.056 0.229Livestock combination 0.020 0.139 0.027 0.162 0.014 0.116 0.011 0.105Vegetable 0.013 0.114 0.012 0.109 0.022 0.148 0.016 0.124Other combination 0.028 0.166 0.023 0.149 0.018 0.131 0.017 0.129Sole proprietorship 0.531 0.499 0.537 0.499 0.180 0.384 0.220 0.415Partnership 0.353 0.478 0.382 0.486 0.328 0.470 0.357 0.479Corporation 0.117 0.321 0.081 0.272 0.491 0.500 0.423 0.494Total land area (1000 ha) 0.319 1.028 0.174 0.439 0.701 1.804 0.799 1.623Total capital ($1,000,000) 0.683 1.085 0.490 0.679 2.815 5.355 2.467 4.535Total sales ($1000) 75.207 69.783 43.153 53.313 1173.786 5289.474 830.942 2895.478Total weeks of hired work 14.623 32.817 7.526 23.124 220.291 761.497 153.166 635.582

Local and regional characteristicsLocal employment growth 7.527 16.108 9.304 16.061 9.640 15.412 8.907 16.132Regional employment growth 6.309 9.034 7.124 8.958 7.785 8.184 6.499 8.892Local specialization 0.191 0.074 0.185 0.062 0.177 0.060 0.184 0.064Regional specialization 0.202 0.046 0.201 0.042 0.190 0.042 0.199 0.045Local unemployment 0.053 0.044 0.054 0.044 0.048 0.038 0.044 0.034Regional unemployment 0.055 0.032 0.055 0.031 0.054 0.031 0.050 0.027

Employment structureLocal manufacturing NRR 0.037 0.034 0.035 0.031 0.046 0.036 0.041 0.035Regional manufacturing NRR 0.035 0.021 0.034 0.019 0.042 0.022 0.037 0.021Local manufacturing LI 0.020 0.033 0.018 0.029 0.027 0.035 0.021 0.029Regional manufacturing LI 0.020 0.024 0.017 0.020 0.027 0.026 0.021 0.022Local manufacturing SB 0.038 0.040 0.039 0.040 0.039 0.037 0.036 0.037Regional manufacturing SB 0.037 0.028 0.036 0.027 0.041 0.028 0.036 0.029Local manufacturing PDSB 0.022 0.024 0.021 0.022 0.027 0.025 0.023 0.023Regional manufacturing PDSB 0.022 0.015 0.021 0.015 0.027 0.017 0.024 0.016Local distributive services 0.109 0.036 0.111 0.034 0.114 0.035 0.113 0.034Regional distributive services 0.109 0.015 0.109 0.014 0.111 0.015 0.110 0.015Local business services 0.069 0.035 0.072 0.034 0.072 0.034 0.068 0.033Regional business services 0.070 0.022 0.072 0.022 0.073 0.023 0.070 0.022Local consumer services 0.202 0.059 0.209 0.056 0.201 0.053 0.198 0.053Regional consumer services 0.204 0.038 0.209 0.038 0.204 0.032 0.201 0.034Local public services 0.209 0.067 0.215 0.065 0.201 0.063 0.203 0.060Regional public services 0.209 0.030 0.212 0.030 0.205 0.030 0.205 0.028

UrbanizationDistance to CA/CMA 56.5 46.7 58.5 50.9 43.7 37.3 54.2 46.6Local population density 43.7 154.9 43.1 144.9 62.1 191.0 50.1 178.8Regional population density 94.0 159.1 100.3 161.6 115.4 180.3 98.5 160.2

Number of operators in the sample 29,533 28,679 9343 2242

Note: these data refer to operators and operator families in private dwellings (residents of collective dwellings are excluded). The data are unweighted to be consistent with theprocedure adopted in the econometric model. The averages for local and regional characteristics should be interpreted as the average local and regional condition experiencedby the operators in the sample. Source: authors’ computations based on Census of Agriculture 2001 and Census of Population 2001 data.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–24 19

Table 4Predicted probability of off-farm labour for selected values and changes of the explanatory variables: full sample (average operator Pr(y¼ 1)¼ 0.412)

Variable Pr(y¼ 1) evaluated at DPr(y¼ 1) Pr(y¼ 1) evaluated at DPr(y¼ 1)

xmin xmax m� (a $ SD) mþ (l $ SD)

Individual characteristicsAge 0.434 0.00 L0.434 0.556 0.379 L0.176Gender 0.429 0.367 L0.062 . . .

Less than high school (omitted)High school 0.368 0.464 0.096 . . .

University certificates 0.404 0.508 0.104 . . .

University degree 0.394 0.601 0.207 . . .

Family characteristicsPresence of children 0.419 0.398 L0.021 . . .

Mobility 0.406 0.446 0.040 . . .

Unpaid housework 0.413 0.410 �0.003 . . .

Unpaid childcare 0.411 0.413 0.001 . . .

Unpaid senior care 0.414 0.388 L0.026 . . .

Farm characteristicsOne operator (omitted)Two operators 0.394 0.432 0.038 . . .

Three operators 0.409 0.439 0.029 . . .

Dairy (omitted)Cattle (beef) 0.282 0.745 0.464 . . .

Hog 0.403 0.667 0.264 . . .

Poultry and egg 0.403 0.780 0.377 . . .

Wheat 0.386 0.830 0.444 . . .

Grain and oilseed (exc. Wheat) 0.323 0.768 0.445 . . .

Field crop (exc. grain & oilseed) 0.371 0.828 0.457 . . .

Fruit 0.398 0.813 0.415 . . .

Miscellaneous specialty 0.344 0.813 0.470 . . .

Livestock combination 0.402 0.823 0.421 . . .

Vegetable 0.407 0.753 0.346 . . .

Other combination 0.402 0.793 0.391 . . .

Sole proprietorship (omitted)Partnership 0.418 0.401 L0.017 . . .

Corporation 0.429 0.328 L0.101 . . .

Area 0.435 1.000 0.565 0.417 0.338 L0.079Capital 0.456 1.000 0.544 0.421 0.308 L0.113Sales 0.415 0.000 L0.415 0.415 0.383 L0.032Hired work 0.420 1.000 0.580 0.417 0.349 L0.068

Local and regional characteristicsLocal employment growth 0.388 0.451 0.063 0.408 0.415 0.007Regional employment growth 0.425 0.401 �0.024 0.414 0.410 �0.004Local specialization 0.416 0.379 �0.036 0.414 0.410 �0.003Regional specialization 0.426 0.364 L0.062 0.417 0.406 L0.011Local unemployment 0.415 0.390 �0.025 0.413 0.411 �0.002Regional unemployment 0.425 0.289 L0.136 0.417 0.407 L0.010

Employment structureLocal manufacturing NRR 0.407 0.460 0.053 0.410 0.414 0.004Regional manufacturing NRR 0.406 0.436 0.030 0.410 0.414 0.004Local manufacturing LI 0.400 0.716 0.316 0.403 0.421 0.019Regional manufacturing LI 0.442 0.183 L0.260 0.430 0.394 L0.037Local manufacturing SB 0.397 0.579 0.182 0.404 0.420 0.016Regional manufacturing SB 0.405 0.435 0.030 0.409 0.415 0.006Local manufacturing PDSB 0.410 0.432 0.023 0.411 0.413 0.003Regional manufacturing PDSB 0.446 0.320 L0.126 0.424 0.400 L0.024Local distributive services 0.387 0.477 0.091 0.408 0.416 0.008Regional distributive services 0.439 0.385 L0.054 0.415 0.409 L0.006Local business services 0.399 0.462 0.063 0.409 0.415 0.006Regional business services 0.353 0.545 0.191 0.399 0.424 0.025Local consumer services 0.368 0.489 0.120 0.406 0.418 0.012Regional consumer services 0.410 0.414 0.003 0.412 0.412 0.001Local public services 0.382 0.473 0.092 0.407 0.417 0.009Regional public services 0.405 0.425 0.019 0.411 0.413 0.003

UrbanizationDistance to CA/CMA 0.391 0.490 0.099 0.404 0.421 0.018Local population density 0.414 0.268 L0.146 0.415 0.408 L0.007Regional population density 0.417 0.318 L0.100 0.416 0.407 L0.009

Note: ‘‘.’’ indicates that the computation is not applicable because the variable is dichotomous. The figures in bold correspond to regression coefficients that are statisticallysignificant at the 90% confidence level or higher. For dichotomous variables, the minimum and maximum values are 0 and 1, respectively. For variables that have a quadraticterm, the level and change of predicted probability include the effect of the quadratic term. For Area, Capital, Sales, and Hired work, a¼ 0.1 and l¼ 0.9; for all the othercontinuous variables a¼ l¼ 0.5. Probabilities are evaluated by holding all other variables at their sample mean. Source: authors’ computations based on estimation results,Census of Agriculture 2001 and Census of Population 2001 data.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–2420

Table 5Predicted probability of off-farm labour for selected values and changes of the explanatory variables: operator of smaller census-farm (average operator Pr(y¼ 1)¼ 0.473)

Variable Pr(y¼ 1) evaluated at DPr(y¼ 1) Pr(y¼ 1) evaluated at DPr(y¼ 1)

xmin xmax m� (a $ SD) m þ (l $ SD)

Individual characteristicsAge 0.484 0.000 L0.484 0.647 0.435 L0.212Gender 0.500 0.401 L0.100 . . .

Less than high school (omitted)High school 0.421 0.534 0.113 . . .

University certificates 0.463 0.590 0.127 . . .

University degree 0.453 0.672 0.219 . . .

Family characteristicsPresence of children 0.472 0.474 0.002 . . .

Mobility 0.472 0.479 0.008 . . .

Unpaid housework 0.489 0.455 L0.034 . . .

Unpaid childcare 0.474 0.471 �0.003 . . .

Unpaid senior care 0.476 0.438 L0.038 . . .

Farm characteristicsOne operator (omitted)Two operators 0.443 0.506 0.063 . . .

Three operators 0.468 0.552 0.084 . . .

Dairy (omitted)Cattle (beef) 0.363 0.708 0.345 . . .

Hog 0.467 0.734 0.266 . . .

Poultry and egg 0.468 0.812 0.344 . . .

Wheat 0.451 0.790 0.340 . . .

Grain and oilseed (exc. wheat) 0.398 0.766 0.368 . . .

Field crop (exc. grain & oilseed) 0.442 0.759 0.317 . . .

Fruit 0.462 0.750 0.288 . . .

Miscellaneous specialty 0.421 0.741 0.320 . . .

Livestock combination 0.465 0.756 0.290 . . .

Vegetable 0.470 0.696 0.226 . . .

Other combination 0.466 0.709 0.243 . . .

Sole proprietorship (omitted)Partnership 0.471 0.476 0.005 . . .

Corporation 0.476 0.443 L0.033 . . .

Area 0.495 0.999 0.504 0.478 0.403 L0.075Capital 0.484 0.846 0.362 0.474 0.456 L0.018Sales 0.638 0.263 L0.375 0.449 0.304 L0.145Hired work 0.477 0.905 0.428 0.473 0.461 L0.012

Local and regional characteristicsLocal employment growth 0.453 0.505 0.051 0.470 0.476 0.006Regional employment growth 0.481 0.466 �0.015 0.474 0.472 �0.002Local specialization 0.479 0.416 �0.064 0.476 0.470 �0.006Regional specialization 0.502 0.374 L0.128 0.484 0.461 L0.023Local unemployment 0.476 0.452 �0.024 0.474 0.472 �0.002Regional unemployment 0.495 0.270 L0.226 0.481 0.464 L0.017

Employment structureLocal manufacturing NRR 0.471 0.493 0.022 0.472 0.474 0.002Regional manufacturing NRR 0.460 0.529 0.069 0.468 0.477 0.009Local manufacturing LI 0.464 0.710 0.246 0.465 0.480 0.015Regional manufacturing LI 0.510 0.169 L0.341 0.496 0.449 L0.047Local manufacturing SB 0.465 0.552 0.087 0.469 0.477 0.008Regional manufacturing SB 0.473 0.472 �0.001 0.473 0.473 0.000Local manufacturing PDSB 0.470 0.500 0.030 0.471 0.474 0.003Regional manufacturing PDSB 0.501 0.390 L0.111 0.483 0.463 L0.020Local distributive services 0.455 0.519 0.064 0.470 0.476 0.006Regional distributive services 0.505 0.440 L0.065 0.476 0.469 L0.007Local business services 0.465 0.501 0.036 0.471 0.475 0.004Regional business services 0.423 0.582 0.159 0.462 0.483 0.021Local consumer services 0.446 0.518 0.072 0.469 0.476 0.008Regional consumer services 0.521 0.409 L0.112 0.482 0.463 L0.019Local public services 0.455 0.508 0.053 0.470 0.475 0.006Regional public services 0.483 0.454 �0.029 0.475 0.471 �0.004

UrbanizationDistance to CA/CMA 0.458 0.546 0.088 0.467 0.479 0.012Local population density 0.474 0.365 �0.109 0.475 0.470 �0.005Regional population density 0.478 0.380 L0.098 0.477 0.468 L0.009

Note: ‘‘.’’ indicates that the computation is not applicable because the variable is dichotomous. The figures in bold correspond to regression coefficients that are statisticallysignificant at the 90% confidence level or higher. For dichotomous variables, the minimum and maximum values are 0 and 1, respectively. For variables that have a quadraticterm, the level and change of predicted probability include the effect of the quadratic term. For Area, Capital, Sales, and Hired work, a¼ 0.1 and l¼ 0.9; for all the othercontinuous variables a¼ l¼ 0.5. Probabilities are evaluated by holding all other variables at their sample mean. Source: authors’ computations based on estimation results,Census of Agriculture 2001 and Census of Population 2001 data.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–24 21

Table 6Predicted probability of off-farm labour for selected values and changes of the explanatory variables: operator of larger census-farm (average operator Pr(y¼ 1)¼ 0.175)

Variable Pr(y¼ 1) evaluated at DPr(y¼ 1) Pr(y¼ 1) evaluated at DPr(y¼ 1)

xmin xmax m� (a $SD) m þ (l $ SD)

Individual characteristicsAge 0.279 0.031 �0.248 0.202 0.157 �0.045Gender 0.161 0.228 0.067 . . .

Less than high school (omitted)High school 0.166 0.185 0.019 . . .

University certificates 0.172 0.204 0.032 . . .

University degree 0.168 0.255 0.087 . . .

Family characteristicsPresence of children 0.177 0.173 �0.004 . . .

Mobility 0.174 0.180 0.006 . . .

Unpaid housework 0.175 0.176 0.001 . . .

Unpaid childcare 0.175 0.175 �0.001 . . .

Unpaid senior care 0.176 0.162 �0.014 . . .

Farm characteristicsOne operator (omitted)Two operators 0.162 0.190 0.028 . . .

Three operators 0.168 0.202 0.033 . . .

Dairy (omitted)Cattle (beef) 0.149 0.364 0.215 . . .

Hog 0.163 0.295 0.132 . . .

Poultry and egg 0.161 0.415 0.254 . . .

Wheat 0.170 0.403 0.233 . . .

Grain and oilseed (exc. wheat) 0.134 0.358 0.225 . . .

Field crop (exc. grain & oilseed) 0.169 0.308 0.139 . . .

Fruit 0.172 0.358 0.185 . . .

Miscellaneous specialty 0.168 0.290 0.121 . . .

Livestock combination 0.174 0.288 0.115 . . .

Vegetable 0.173 0.278 0.105 . . .

Other combination 0.172 0.365 0.193 . . .

Sole proprietorship (omitted)Partnership 0.179 0.166 �0.013 . . .

Corporation 0.190 0.159 L0.031 . . .

Area 0.177 0.261 0.084 0.175 0.169 �0.006Capital 0.187 0.595 0.408 0.176 0.158 L0.022Sales 0.177 0.010 �0.167 0.176 0.163 �0.013Hired work 0.178 0.719 0.541 0.175 0.163 L0.012

Local and regional characteristicsLocal employment growth 0.141 0.221 0.080 0.170 0.180 0.010Regional employment growth 0.187 0.166 �0.021 0.176 0.174 �0.003Local specialization 0.182 0.116 �0.066 0.178 0.172 �0.006Regional specialization 0.178 0.164 �0.013 0.176 0.174 �0.002Local unemployment 0.196 0.062 L0.134 0.183 0.167 L0.016Regional unemployment 0.161 0.353 0.192 0.170 0.181 0.011

Employment structureLocal manufacturing NRR 0.169 0.219 0.050 0.172 0.178 0.005Regional manufacturing NRR 0.188 0.135 �0.053 0.179 0.171 �0.008Local manufacturing LI 0.170 0.250 0.080 0.172 0.178 0.007Regional manufacturing LI 0.198 0.068 L0.130 0.187 0.164 L0.023Local manufacturing SB 0.170 0.233 0.063 0.173 0.177 0.005Regional manufacturing SB 0.168 0.199 0.031 0.172 0.178 0.006Local manufacturing PDSB 0.185 0.113 L0.072 0.180 0.170 L0.009Regional manufacturing PDSB 0.168 0.189 0.021 0.173 0.177 0.005Local distributive services 0.165 0.202 0.038 0.173 0.177 0.003Regional distributive services 0.208 0.146 �0.062 0.179 0.171 �0.007Local business services 0.215 0.091 L0.124 0.184 0.166 L0.018Regional business services 0.144 0.267 0.124 0.167 0.183 0.016Local consumer services 0.161 0.197 0.036 0.173 0.177 0.004Regional consumer services 0.178 0.172 �0.006 0.175 0.175 �0.001Local public services 0.161 0.206 0.045 0.173 0.177 0.004Regional public services 0.183 0.158 �0.025 0.177 0.173 �0.003

UrbanizationDistance to CA/CMA 0.159 0.421 0.262 0.167 0.181 0.014Local population density 0.173 0.291 0.118 0.172 0.178 0.005Regional population density 0.178 0.129 �0.050 0.178 0.172 �0.005

Note: ‘‘.’’ indicates that the computation is not applicable because the variable is dichotomous. The figures in bold correspond to regression coefficients that are statisticallysignificant at the 90% confidence level or higher. For dichotomous variables, the minimum and maximum values are 0 and 1, respectively. For variables that have a quadraticterm, the level and change of predicted probability include the effect of the quadratic term. For Area, Capital, Sales, and Hired work, a¼ 0.1 and l¼ 0.9; for all the othercontinuous variables a¼ l¼ 0.5. Probabilities are evaluated by holding all other variables at their sample mean. Source: authors’ computations based on estimation results,Census of Agriculture 2001 and Census of Population 2001 data.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–2422

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–24 23

a census farm near a city are more likely to report off-farm work,both because farming individuals could access city jobs andbecause city workers could access a hobby farm. However, thisexpectation neglects the nature of urbanization in Canada and itseffects on the value of on-farm work. Firstly, Canadian cities andtowns, generally, were founded on the best agricultural land.Operators of a census-farm living at a distance from a city, ingeneral, live on marginal agricultural land. In addition, recent re-search suggests that urbanization increases net returns to agricul-ture due to the combined effect of a reduction in transportationcosts, lower cost structures, and potentially higher output prices(Livanis et al., 2006). As a result, the value of on-farm work can behigher than the value of off-farm work for operators who live closerto cities. Consequently, individuals operating a census-farm inmore rural settings, with lower agricultural returns, are more likelyto combine farming with one or more employment pursuits off thefarm.

5. Conclusions

In this paper we investigated the determinants of off-farmemployment for operators associated with different sizes of agri-cultural holdings. By combining micro-level data with communitylevel data, this analysis offers an assessment of the relative im-portance of individual, farm, community and regional factors on theoff-farm participation decision.

Overall, it appears that human capital and farm characteristicsare significant and have similar effects for operators of both smallerand larger holdings. Among the individual-level covariates, edu-cational attainment is a major determinant of the ability of a farmerto participate in off-farm work. However, the type of farm must beamenable to off-farm work (i.e. dairy is not). Increasing the numberof operators within a farm also increases the likelihood of off-farmemployment by one of the operators. Hired work is a complementto farm labour by the operator as it allows the farmer to specializeand increase the marginal value of farm labour.

In contrast, family characteristics, as well as local and regionalcharacteristics, appear to be more important for smaller operations.Local labor market characteristics do affect the likelihood of off-farm employment, but much more so for operators of smallerfarms. This finding is consistent with the prevailing literature onsmaller agricultural holdings, which emphasizes the combinationof family, community and regional factors in determining familywell-being (Fuller and Bollman, 1992). In particular, the importanceof pluriactivity by members of families associated with agriculturalholdings has been shown to be relevant in Canada, where, in ad-dition to the process of exit from the agricultural sector, a partialadjustment of agricultural labour through participation in off-farmlabour has become a pervasive strategy of census-farm operatorsand their families (Olfert and Stabler, 1994). These trends havebrought recognition to the fact that, in many rural areas, pluri-activity is likely to be an important strategy for maintaininghousehold livelihoods (OECD, 1995). Moreover, it has been ob-served that part-time farming is a relatively unimportant facilitatorof entry into commercial farming (Bollman and Steeves, 1982). Inother words, off-farm employment is not part of a phase-in periodto full-time farming. The implication is that to maximize familyincome, operators of agricultural holdings may have to face thechoice of getting bigger or getting smaller (Bollman, 1991), and foroperators of smaller holdings, the combination of family and localmarket conditions can play a critical role in determining the well-being of their family.

An important result is the unexpected effect of proximity tourban areas. The results from this analysis suggest that urban re-gions are not the core labour markets for the operators who areinvolved in off-farm labour; rather urban areas appear to enhance

the value of the operator’s labour on the holding. This result sug-gests some degree of disjunction between the off-farm laboursupply of farm operators and the urban economy. This finding hasimplications for policies aimed to enhance the incomes of familiesassociated with farming. There is evidence that a strong agricul-tural sector is neither necessary nor sufficient for high and growinghousehold income in rural areas; research findings from the USindicate that what matters is the linkage of farm factor markets,primarily labour, with the non-farm sector (Gardner, 2005; Fullerand Bollman, 1992). Hence, off-farm labour represents one of theprimary chains of transmission of wealth between the farm andnon-farm sector. The results of this analysis, however, suggest thatthis chain of transmission from urban to farm operator householdsmay be via the demand for farm products for sale in the urban la-bour markets.

With a growing share of employment opportunities concen-trated in predominantly urbanised regions, the linkages betweenurban labour markets and rural populations remain vital for theeconomic sustainability of rural areas (Bollman and Biggs, 1992;Schindegger and Krajasits, 1997). However, the present analysissuggests that, for farm operators involved in off-farm work, themain linkages are with the rural labour market itself. This findingparallels research suggesting commuting flows are not alwaysunidirectional and not even exclusively dominated by in-com-muting in major urban centres (Green and Meyer, 1997). Researchconducted in Southern Ontario, for instance, indicates a decliningurban influence, as a pole of attraction, in the face of an increasingrural ‘‘self-sufficiency’’ or a rural-to-rural pattern of interaction(Green and Meyer, 1997). The nature of this rural-to-rural in-teraction should be further investigated with a specific focus on itsrelevance to off-farm labour opportunities for members of familiesassociated with smaller agricultural holdings.

References

Anselin, L., 1988. Spatial Econometrics: Methods and Models. Kluwer AcademicsPublisher, Dordrecht, The Netherlands.

Benjamin, C., Corsi, A., Guyomard, H., 1996. Modelling labour decision of Frenchagricultural households. Applied Economics 28, 1577–1589.

Boarnet, M.G., Chalermpong, S., Geho, E., 2005. Specification issues in models ofpopulation and employment growth. Papers in Regional Science 84 (1), 21–46.

Bollman, R.D., 1979. Off-farm work by farmers: an application of the kinked demandcurve of labour. Canadian Journal of Agricultural Economics 27 (1), 37–60.

Bollman, R.D., 1991. Efficiency aspects of part-time garming. In: Hallberg, M.C.,Findeis, J.L., Lass, D.A. (Eds.), Multiple Jobholding Among Farm Families. IowaState University Press, Ames, pp. 112–139.

Bollman, R.D., Biggs, B., 1992. Rural and small town Canada: an overview. In:Bollman, R.D. (Ed.), Rural and Small Town Canada. Thompson EducationalPublishing, Toronto.

Bollman, R.D., Steeves, A.D., 1982. The stocks and flows of Canadian census-farmoperators over the period 1996–1976. The Canadian Review of Sociology andAnthropology 19 (4), 576–590.

Bollman, R.D., Fuller, A.M., Ehrensaft, P., 1992. Rural jobs: trends and opportunities.Canadian Journal of Agricultural Economics 40 (4), 605–622.

Findeis, J.L., Jensen, L., 1998. Employment opportunities in rural areas: implicationsfor poverty in a changing policy environment. American Journal of AgriculturalEconomics 80 (5), 1000–1007.

Findeis, J.L., Lass, D.A., Hallberg, M.C., 1991. Effects of location on off-farm em-ployment decision. In: Hallberg, M.C., Findeis, J.L., Lass, D.A. (Eds.), MultipleJobholding Among Farm Families. Iowa State University Press, Ames, pp.263–276.

Fuller, A.M., Bollman, R.D., 1992. Farm family linkages to the non-farm sector: therole of off-farm income of farm families. In: Bollman, R.D. (Ed.), Rural and SmallTown Canada. Thompson Educational Publishing, Toronto.

Gardner, B.L., 2005. Causes of Rural Economic Development. In Reshaping Agri-culture’s Contribution to Society. Proceedings of the 25th International Con-ference of Agricultural Economics. Blackwell.

Glaeser, E., Mare, D., 2001. Cities and skills. Journal of Labor Economics 19 (1),316–342.

Green, M.B., Meyer, S.P., 1997. Occupational stratification of rural commuting. In:Bollman, R.D., Bryden, J.M. (Eds.), Rural Employment: An International Per-spective. CAB International, New York.

Howard, W., Swidersky, M., 2000. Estimating the off-farm labor supply in Canada.Canadian Journal of Agricultural Economics 48 (1), 1–14.

A. Alasia et al. / Journal of Rural Studies 25 (2009) 12–2424

Huffman, W.E., 1991. Agricultural household models: survey and critique. In:Hallberg, M.C., Findeis, J.L., Lass, D.A. (Eds.), Multiple Job Holding Among FarmFamilies. Iowa State University Press, Ames.

Livanis, G., Moss, C.B., Breneman, V.E., Nehring, R.F., 2006. Urban sprawl andfarmland prices. American Journal of Agricultural Economics 88 (4), 915–929.

Long, S.J., Freese, J., 2001. Regression Models for Categorical Dependent VariablesUsing Stata. Stata Press, College Station, Texas.

OECD, 1995. Creating Employment for Rural Development – New Policy Approaches.OECD, Paris.

Olfert, M.R., Stabler, J.C., 1994. Industrial restructuring of prairie labour force: spatialand gender impacts. Canadian Journal of Regional Science XVII (2), 133–152.

Phimister, E., Vera-Toscano, E., Weersink, A., 2002. Female participation and labormarket attachment in rural Canada. American Journal of Agricultural Economics84 (1), 210–221.

Rosen, S., 1978. Substitution and division of labor. Economica 45 (2), 235–250.

Schindegger, F., Krajasits, C., 1997. Commuting: its importance for rural employmentanalysis. In: Bollman, R.D., Bryden, J.M. (Eds.), Rural Employment: An In-ternational Perspective. CAB International, New York.

Statistics Canada, 2002. Farm Operator Data for the 2001 Census of Agriculture.Catalogue no. 95F0355XIE. Statistics Canada, Ottawa.

Vera-Toscano, E., Weersink, A., Phimister, E., 2003. Earnings mobility of rural vsurban workers in Canada. Canadian Journal of Agricultural Economics 51 (2),241–258.

Vera-Toscano, E., Phimister, E., Weersink, A., 2004. Panel estimates of the Canadianrural/urban women’s wage gap. American Journal of Agricultural Economics 86(4), 1138–1151.

Weersink, A., 1992. Off-farm labour decisions by Ontario Swine Producers. CanadianJournal of Agricultural Economics 40 (2), 235–252.

Wheeler, C.H., 2001. Search, sorting and urban agglomeration. Journal of LaborEconomics 19 (4), 879–899.