Data and definitions

Box 1  Key definitions

A census farm is defined as "a farm, ranch or other agricultural operation which produces at least one of the following products intended for sale: crops, livestock, poultry, animal products, greenhouse or nursery products, Christmas trees, mushrooms, sod, honey or bees, and maple syrup products." (Statistics Canada 2003b:6).

A smaller census farm is defined as an operation reporting total gross farm receipts of less than $250,000 for the census year. In 2001, there were 290,510 operators associated with a smaller census farm.

A larger census farm is defined as an operation reporting total gross farm receipts equal or greater than $250,000 for the census year. In 2001, there were 55,685 operators associated with a larger census farm.

Note that a typical farm has net cash revenue (before depreciation) that is about 15% of gross farm revenues (Statistics Canada 2000). Thus, a typical census farm with gross revenue of less than $250,000 would generate net revenue (before depreciation) of less than $37,500. Depreciation is typically one-half of net revenue before depreciation. Thus, after depreciation is taken into account, this income would be below the low income cut-off for a rural family of four ($23,713 in 2000). Thus, census farms with gross revenue less than $250,000 are designated as smaller census farms. We acknowledge there is a wide variability of net farm revenue as a percent of gross farm revenue when one compares farms specializing in different enterprises and, also, different operators have widely varying degrees of good luck and good management.

Farm operator "refers to those persons responsible for the day-to-day management decisions made in the operation of the census farm." (Statistics Canada 2003b:7).

Community: In this bulletin, "community" is defined as a Census Consolidated Subdivision (CCS). The two terms, community and CCS, are used synonymously. CCSs are defined according to the 1996 census geography. It should be mentioned that Statistics Canada does not provide a standard definition for the term community. The term is generically used to refer to administrative and statistical geographic units of smaller spatial areas and at an intermediate level between the provincial or regional level (economic regions, health region, Census agricultural regions, etc.) and the micro-geographic levels (Dissemination Areas, blocks or neighbourhoods).

A Census Consolidated Subdivision (CCS) is a grouping of two or more Census Subdivisions (CSDs), where a CSD is an incorporated town or municipality. For the detailed definition, see Statistics Canada (2003b). A typical case is where an incorporated town and the surrounding incorporated rural municipality have been consolidated as a CCS for statistical purposes.

Larger urban centre:This term is used to identify a Census Metropolitan Area (CMA) or Census Agglomeration (CA) of any size. In 2001, a CMA had an urban core of 100,000 or more and a CA has an urban core of 10,000 to 99,999. For the detailed definition, see Statistics Canada (2003b).  It should be noted that the CMA and CA definitions were modified in 2006 to: a census metropolitan area must have a total population of at least 100,000 of which 50,000 or more live in the urban core. A census agglomeration must have an urban core population of at least 10,000 (Statistics Canada 2007a).

Region: Similar to the situation for "community", the term "region" also does not have a unique definition when  used by Statistics Canada. For each community, we define its region, or regional milieu, as a set of communities within a certain distance of the given community. To measure the characteristics of this region, for a given variable, we use a spatially lagged variable computed from the community indicator. Specifically, a spatially lagged variable for each community is computed as the weighted average of the value of the indicator in the surrounding communities, where the weights are the inverse of the squared distance between community centroids (for details, see Alasia et al. 2007).

Box 2  Data source

This study used data from the 2006 Census of Agriculture to update the information on trends. The inter-relationships were estimated using data from the 2001 Agriculture-Population Linkage database combined with community-level data from the  2001 Census of Population. This database includes 70,851 census farm operators, of which we excluded 513 operators in collective dwellings (which would largely be residents in Hutterite Colonies). Community level data are obtained from the 2001 Census of Population, tabulated for constant 1996 census geography; this corresponds to 2,607 CCSs.  

For the variables used in this research, data were available for 2,382 CCSs (1996 census geography) because community data could not be tabulated for CCSs with population less than 250 individuals for data quality and confidentiality reasons. The combination of the Agriculture-Population Linkage database and community level database yielded the data set used for estimation, which encompassed 69,797 farm operators and 1,746 CCSs out of 1,783 CCSs (2001 census geography) for which Census of Agriculture data was collected.

The Agriculture-Population Linkage database for 2006 was released in December, 2008 and was not available for the analysis reported in this bulletin.

Box 3  Methodology

The results presented in this bulletin are based on an econometric model (specifically a probit model) that is derived from a theoretical model of farm labour allocation decisions (farm household model). The econometric model that is estimated can be summarized as follows:

Pr(M = 1 | x) = Φ(β₁H + β₂Z + β₃K + β₄R)

where the probability of observing off-farm work, Pr(M=1), is a function of individual (H), family (Z), farm (K) and community and regional labour market characteristics (R), the β's represent the coefficients to be estimated,  and Φ(.) denotes the cumulative normal distribution function. We use a total of 60 variables to capture these effects. A description of the variables used in the model is provided in Appendix Table A.2.

It should be emphasized that we identify two components of the regional milieu effect: the community effect and the regional effect. This distinction is introduced by means of CCS variables to capture the effect of community characteristics and their corresponding spatial lag to capture the regional effect. For each CCS and indicator of interest, the spatial lag is a distance weighted average of the neighbouring CCSs' values for that given indicator (see also the definition of geography in Box 1).

The probit model is estimated for the whole sample of operators and two sub-samples, corresponding to operators associated with smaller and larger census farms. We define operators of smaller census farms as those reporting total gross farm receipts of less than $250,000 for the census year. This sub-sample corresponds to 58,212 operators (83% of the total sample). Operators of larger census farms are defined as those with total gross farm receipts equal to or greater than $250,000 for the census year; this corresponds to 11,585 operators (17% of the total sample).

The results of these models are then used to compute the predicted probabilities of off-farm work, by using the estimated coefficients (β) and a specific set of values of the explanatory variables H, Z, K, and R. The specific set of values of the explanatory variables is intended to represent typical profiles of census farm operators. These predicted probabilities are often compared with the predicted probability of the average operator of a census farm (for three cases: whole sample, smaller and larger census farms). An average operator is an operator with sample average values for each of the explanatory variables used in the model. For instance we compare the predicted probability of the average operator with the probability of the operator with a university degree. Hence, we consider the case of an operator who has all the "average" characteristics (average value for his sample) and we compare it to the same operator whose only difference is to have a university degree.

We also compare the difference in the predicted probability of off-farm work for observations that differ by one standard deviation to get a feel for how off-farm work participation varies across the population of census farm operators. A standard deviation is a measure of spread of the data. Typically, 68% of the observations are within plus or minus one standard deviation of the average. For instance, with a 13 year standard deviation for the distribution of operators by age, we are saying that about 68% of the operators are within plus or minus 13 years of the age of the average operator.

For more details on the methodology see Alasia, Bollman, Weersink and Cranfield (2007).