Dependent Variables

Four farm categories can be defined using CL-CEAG for a given census using information of past, current, and future census periods: (1) a Continuing: a farm in the previous (Farm_{t -1}), current (Farm_t), and next census (Farm_{t +1}), (Farm_{t +1}|Farm_t|Farm_{t -1}); (2) an Entry-Stay: a farm not in the previous census, but in the current and next one (Farm_{t +1}|Farm_t|NoFarm_{t -1}); (3) an Old Exit: a farm in the previous and current census, but not in the next one (NoFarm_{t +1}|Farm_t|Farm_{t -1}); and (4) an Entry-Exit: a one-census farm (NoFarm_{t +1}|Farm_t|NoFarm_{t -1}) (see Table 1). An Entry-Exit farm qualifies both as an entrant farm and an exit farm in a given census year. Thus, entry farms (Entrant) are the farms that appear for the first time in the census database in a given census year (Entry-Stay plus Entry-Exit), while exit farms (Exit) are the farms that appear for the last time in the census database in a given census year (Old Exit plus Entry-Exit).

Table 1.

Categorization of farms in a given census based on status in the prior, current, and future census periods

^a Farm—farm operated in the associated census,

^b NoFarm—farm did not operate in the associated census.

Over the last 25 years, the number of farms in the latest census has fallen by approximately one-third from 280,043 in 1991 to 193,492 census farms in 2016 (see Table 2). To illustrate the four categories of entry/exit farms in relation to the values reported in the census, note that the number of farms in census t (Farm Numbers_t) is given by

(1)$$\eqalign{Farm\,Numbers_t& = Farm\,Numbers_{t-1}-( Old\,Exit_{t-1} + EntryExit_{t-1}) \cr & \quad + EntryStay_t + EntryExit_t} $$

For example, the 205,730 farms in the 2011 census year is equal to

(2)$$\eqalign{Farm\,Numbers_{2011}& = Farm\,Numbers_{2006}-( {Old\,Exit_{2006} + EntryExit_{2006}} ) \cr & \quad + EntryStay_{2011} + EntryExit_{2011}} $$

$$205, \;730 = 229, \;373\ndash ( 40, \;519 + 16, \;612) + 16, \;759 + 16, \;729$$

There were 33,488 new farms in the 2011 census (=16,759 (EntryStay ₂₀₁₁) + 16,729 (EntryExit ₂₀₁₁)), but 57,131 farms from the 229,373 in the 2006 census exited the sector (=40,519 (Old Exit ₂₀₀₆) + 16,612 (EntryExit ₂₀₀₆)), resulting in a net decline of 23,643 farms between the 2006 to 2011 census periods.

Table 2.

Continuing, Old Exit, Entry-Stay, Entry-Exit, and total farm numbers in Canada, 1991–2016

^a Continuing are farms in the previous census (t−1), the current one (t), and the next one (t+1).

^b Old Exit are farms in the previous census (t−1) and current census (t), but not in the next one (t+1).

^c Entry-Stay are farms not in the previous census (t−1), but in the current (t) and the next census (t+1).

^d Entry-Exit are farms not in the previous census (t−1), but in the current one (t) and not in the next one (t+1).

^e It is not known which of the farms in the 2011 and 2016 census will exit the sector by the next census in 2021.

^f The values represent the number of new entrants in total. It cannot be broken up into EntryStay and EntryExit until the next census in 2021.

It is not known which of the 193,492 farms in the 2016 census will have exited farming by the time of the next census in 2021. As a result, we are unable to determine the values for Continuing ₂₀₁₆, Old Exit ₂₀₁₆, EntryStay ₂₀₁₆, and EntryExit ₂₀₁₆. Although the two types of entrants are unobservable, we do know the total number of entrants and are, thus, able to calculate the number of the farms in the 2016 census through [1] as

$$\eqalign{Farm\,Numbers_{2016}& = Farm\,Numbers_{2011}\ndash ( Old\,Exit_{2011} + EntryExit_{2011}) \cr & \quad + Entrant_{2016}} \eqno ( 1 \prime \;) $$

$$193, \;492 = 205, \;730\ndash ( 70, \;852 + 16, \;730) + 58, \;614$$

Note that the percentages listed in Table 2 for each of the four farm categories do not total to 100 since the sum of all farms in a census period is not the simple sum of the number of farms in each category but rather covers both the current and the previous census periods as defined above.

The number of new entrants fell by almost 60 percent from 1991 to 2011 (83,079 to 33,488), although the number jumped significantly in the last census to 58,614. The number of exits has also fallen over time due partially to the lower number of the total farmers, but generally between one-quarter to one-third of census farms exit the sector before the next census. A significant portion of entrants into agriculture are not in operation by the time of the next census (Entry-Exit in Table 1). For example, half (16,729) of the 33,488 farms who started farming between the 2006 and 2011 census had left farming by the 2016 census.

Explanatory Variables

Kimhi and Bollman (Reference Kimhi and Bollman1999) and Kimhi (Reference Kimhi2000) develop a model using neoclassical human capital theory in which an individual weighs the present value of utility of either exiting (entering) agriculture as compared with the present value of utility from staying in (out of) farming. Based on this conceptual model and previous literature, the set of independent variables on exit decisions is categorized into three groups: operator characteristics (O), farm characteristics (F), and regional market characteristics (R). Definitions for the explanatory variables within each category are given in Table 3 along with the associated average value by type of farm defined by the four categories of farm groups.

Table 3.

Definition and average value (or percentage of farm numbers) of explanatory variables by farm group, 1991–2016

Operator characteristics include variables for age, gender, and multigenerational farming. Their hypothesized relationships with the outcomes vary between the entry and the exit decisions. Similar to other types of businesses, people tend to enter farming when they are young (Hoppe and Korb Reference Hoppe and Korb2006). While younger individuals are more likely to enter farming, the relationship between age and exiting is hypothesized to be ambiguous. Farmers will eventually retire from farming, so on the one hand, the likelihood of exit would be expected to increase with age. However, many farmers can remain as part of an operation well past the normal retirement age (Gale Reference Gale2003), and a high proportion of new entrants, hypothesized to be younger generally, will drop out of farming. In addition, Katchova and Ahearn (Reference Katchova and Ahearn2016) find that younger new entrants are more likely to operate larger farms and likely to grow over time, which may put them at more financial risk. Thus, the sign of the correlation between age and exit is expected to be ambiguous. There are seven categorical age variables defined in CL-CEAG (see Table 3) that allows for three age groupings for single or multiple operators of the farm.

More women are involved in farming than in previous generations due to changing attitudes (Ahearn and Newton Reference Ahearn and Newton2009). Thus, female operators are assumed to be more likely to enter farming over time ceteris paribus (not that more females are expected to enter than males). Similar to age, the relationship between gender and exit is expected to be ambiguous. Multigenerational operations are assumed to have higher entry rates as the adjustment cost of entry is lower and the farming returns to established units are higher than for new operations. For the same reasons, retention rates for multigenerational farms are assumed to be higher and the exit rates consequently lower (Dong et al. Reference Dong, Hennessy, Jensen and Volpe2016).

Higher net farm income is hypothesized to increase the entry rate and slow the exodus from farming (i.e., von Massow et al. Reference Von Massow, Weersink and Turvey1992; Gale Reference Gale1993; Mishra, Fannin, and Joo Reference Mishra, Fannin and Joo2014). Farm characteristic variables include farm size (farmland area), capital, sales, and farm type. It is assumed that new operations are more likely to be smaller and focused on the production of goods with lower adjustment costs. Several studies, including Mishra, Fannin, and Joo (Reference Mishra, Fannin and Joo2014), have found that part-time farming eases the entry into and out of farming. Farm types with lower adjustment costs that increase the rate of entry are also likely to slow the rate of exit. For example, crop farms tend to be more amenable to entry through the ability to work both on and off the farm due to fewer labor requirements than livestock operations. In addition to labor requirements, the entry may be lower in sectors with high barriers to entry in the form of quotas (i.e., poultry and dairy). Farm size and sales are indicators of farm profitability and potentially technical efficiency and are, thus, assumed to be negatively correlated with exit (Kimhi Reference Kimhi2000; Dong et al. Reference Dong, Hennessy, Jensen and Volpe2016).

In addition to regional dummy variables, the regional market characteristics consist of two other variables: the population density and the median income of the CCS in which the operation is located. While population density may increase farmland values and, thus, the adjustment costs of entry, it may also increase the demand for local produce that can be met by small operations selling through local farmer markets (Govindasamy, Hossain, and Adelaja Reference Govindasamy, Hossain and Adelaja1999). While this may have a positive effect on entry, off-farm employment opportunities and the associated wage rates are higher in more densely populated regions, thereby increasing the relative returns to staying out of farming. Thus, the expected sign of the relationship between population density of the CCS and entry is ambiguous. Similarly, its relationship with exit is unknown a priori. The median farm income is an indicator of the importance of agriculture within the CCS. As with farm price, we expect higher farm incomes increase the relative attractiveness of farming and, thus, increase (decrease) the likelihood of entry (exit).

Econometric Model

A linear probability model (LPM) is used to assess the determinants of the outcome variables in two scenarios: (1) farms that exit between the current and next census (Old Exit) compared with those not exiting (Continuing) conditional on both groups farming in the previous census or NoFarm_{t +1}|Farm_t|Farm_{t -1} versus Farm_{t +1}|Farm_t|Farm_{t -1}, and (2) new entrants that exit (Entry-Exits) compared with new entrants that continue farming into the next census (Entry-Stay) or NoFarm_{t +1}|Farm_t|NoFarm_{t -1} versus Farm_{t +1}|Farm_t|NoFarm_{t -1}.

To illustrate the model for the two scenarios, we begin with the specification of the dependent variable for the exiting farm analysis for scenario (1) on exiting versus continuing farms conditional on both farming in the previous census.

(3)$$Old\,Exit_{it} = \left\{{\matrix{ {1\;\;{\rm if}\,{\rm farm}\,{\rm is}\,{\rm in}\,{\rm operation}\,{\rm in}\,{\rm census}\,{\rm period}\,t-1\,{\rm and}\,t\,{\rm but}\,{\rm not}\,{\rm in}\,t + 1} \cr {0\;\;{\rm if}\,{\rm farm}\,{\rm is}\,{\rm in}\,{\rm operation}\,{\rm in}\,{\rm census}\,{\rm periods}\,t-1, \;\;\,t, \;\;\;{\rm and}\,t + 1\;( {{\rm Continuing}} ) } \cr } } \right.$$

The LPM, which is an ordinary least squares model with a binary dependent variable, is defined as follows:

(4)$$Old\,Exit_{it} = {\rm \gamma }_0 + {\rm \delta }_OO_{it} + {\rm \delta }_FF_{it} + {\rm \delta }_RR_{it} + {\rm \delta }_TT + {\rm \varepsilon }$$

where γ₀ represents the intercept and δ's are the vectors of parameters associated with the vectors of explanatory variables, O, F, and R that are defined in the next section. T is a vector of indicators for census years. Lastly, ɛ is the robust standard error with mean zero.

Similarly, the dependent variable for the other regression model that compares the two types of entrants is defined as

(5)$$EntryExit_{it} = \left\{{\matrix{ {1\;{\rm if}\,{\rm operation}\,{\rm enters}\,{\rm in}\,t\,{\rm and}\,{\rm exits}\,{\rm in}\,t + 1\;( {{\rm Entry}\,{\rm Exit}} ) } \cr {0\;{\rm if}\,{\rm operation}\,{\rm enters}\,{\rm in}\,t\,{\rm and}\,{\rm continues}\,{\rm in}\,t + 1\;( {{\rm Entry}\,{\rm Stay}} ) } \cr } } \right.$$

The LPM is used as the main model given its ease of interpretation and that the predicted probabilities of Old Exit and Entry-Exit lie well within the unit interval, which is an important condition for the unbiasedness of the LPM (Horrace and Oaxaca Reference Horrace and Oaxaca2006). Given the binary dependent variables, we show results from the logit model in the Appendix for robustness.

Note that each of the two regressions are conditional regressions that rely on specific subsamples of the data set. The Old Exit regression is based on the subset of farms that existed in the previous census period, while the Entry-Exit regression is based on the subset of farms that did not exist in the previous census period t−1 and started to appear from census period t. This approach of focusing on exit decisions and defining exiting farms as opposed to continuing farms is based on the nature of the census data that do not include information on the potential entrants and is consistent with the approach taken by previous studies by Ahearn and Newton (Reference Ahearn and Newton2009) and Katchova and Ahearn (Reference Katchova and Ahearn2017).

Summary Statistics of Explanatory Variables by Entry/Exit Category

An overview of the independent variables used in the analysis is summarized by farm groups (Continuing, Entry-Stay, Old Exit, and Entry-Exit) in Table 3. The independent variables are grouped into three categories: operator characteristics (O), farm characteristics (F), and regional characteristics (R).