Publications

    Rural and Small Town Canada Analysis Bulletin

    Employment shifts in natural resource sectors: A focus on rural value chains

    Data and definitions

    Warning View the most recent version.

    Archived Content

    Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.

    Box 1
    Geographic level of analysis

    In this bulletin two types of geographies are used: census divisions and regional types.
    A census division (CD) is the general term for provincially legislated areas (such as county, municipalité régionale de comté and regional district) or their equivalents. Census divisions are intermediate geographic areas between the province/territory level and the municipality (census subdivision).

    In our study, a census division (CD) represents a region which is the basic geographical building block from which regional types are generated. As CD boundaries change from census to census, our data are tabulated using the constant geographic boundaries of 1996. There were 288 CDs in 1996.

    Regional type is defined using the Organisation for Economic Co-operation and Development (OECD) typology. An OECDrural community is a community with a population density less than 150 people per square kilometre. Using this definition of rural community, the following classification of regions is used by the OECD:

    • Predominantly rural regions: more than 50% of the population lives in a 'rural community';
    • Intermediate regions: 15% to 49% of the population lives in a 'rural community'; and
    • Predominantly urban regions: less than 15% of the population lives in a 'rural community'.

    Predominantly rural regions are further broken down to recognize diversity within predominantly rural regions. There are three types of predominantly rural regions: rural metro-adjacent regions, rural non-metro-adjacent regions and rural northern regions (Ehrensaft and Beeman, 1992).

    Rural and Small Town Canada Analysis Bulletins address issues of interest to rural Canada such as employment trends, education levels, health status, Internet usage and number of firms by type, among others.

    As discussed in Puderer (2009) and du Plessis et al. (2001), there are numerous possible operational definitions of urban and rural areas. In this bulletin, we are using the definition that we judge to be the most appropriate for this analysis.

    End of text box 1

    Box 2
    Data source and definition of the value chain of resource sectors

    The employment data used in this analysis are from the Censuses of Population for 1991, 1996 and 2001 and refer to the experienced labour force for both genders 15 years of age and over. The experienced labour force refers to individuals who were employed during the week prior to census day plus those who were unemployed but who had last worked for pay or in self-employment since January 1st of the previous year.

     The delineation of value chains used in this analysis is based on the 1980 Standard Industrial Classification - Establishments (hereafter referred to as SIC-E) (Statistics Canada 1980).

    For the purpose of this research, we defined comparable value chain structures for each resource sector. Five natural resource sectors are considered: (1) agriculture; (2) the fishery; (3) forestry; (4) mining; and (5) energy. Appendix Table A.1 shows the comparable components of the value chain of each resource sector together with the corresponding industry codes for each component in each value chain.

    We use Standard Industrial Classification (SIC-E) codes, instead of the more recent North American Industry Classification System (NAICS) codes (Statistics Canada 2003), because only the SIC-E codes can be applied to the census data for 1991, 1996, and 2001. This provides a better historical perspective (a similar analysis can now be carried out by comparing 2001 and 2006 using the NAICS classification). Employment data on the Census of Population was coded at the 3-digit level of the SIC-E. Thus, our classification is constrained to the 3-digit level. The entire set of SIC-E codes available for use in this analysis is available online on the Statistics Canada web page (Statistics Canada 1980).

    The focus of our analysis is on primary production, services to primary production, wholesaling and the first-stage of processing (i.e. manufacturing). Beyond the first-stage of processing, it becomes increasingly difficult to assign an industry to a specific value chain. Based on the detailed description of the SIC-E codes (Statistics Canada 1980), we assigned industries that appear to have reasonably strong linkages with the natural resource sector. For some industry codes, the Statistics Canada Input-Output Tables 2000 were used to ascertain major industry linkages along the production chain. Although we did not use fixed thresholds, the Input-Output tables provided enough insight to determine major linkages in order to assign the SIC-E code to a specific value chain.

    We acknowledge that the operational delineation of a value chain, in general, and that based on an industry code system, in particular, remains challenging. First, it is conceptually difficult to delimit the chain because the sectoral linkages from one stage to the next stage of processing become increasingly complex as we move to higher value-added activities. There are a number of activities where the association with a specific sector is difficult if not impossible to assign (this problem is typical of some service sectors such as insurance, transportation, packaging, etc). An insurance establishment or a trucking establishment may have customers associated with different value chains. Thus, these establishments, generally, can not be allocated to a given value chain. The complexity and substitutability of production inputs also increases as we move away from the primary production stage, which makes it difficult to assign an establishment to a specific value chain. Second, the demarcation of a value chain is problematic because of the way industry statistics are reported. For example, the standard industrial classification does not always overlap with a specific sectoral chain. As a result of these challenges, the operational definition of a value chain, even if in broad sectoral terms, implies a certain degree of simplification.

    In spite of these limitations, the classification presented in Appendix Table A.1 represents our attempt to develop a framework that allows some comparability across major natural resource sector and permits an analysis of their changes across space and time.

    End of text box 2

    Box 3
    Methodology: Measuring spatial association

    To assess the spatial distribution of value chain segments we use a set of indicators and analytical methods. All these methods rely on the definition of spatial interactions as captured by a spatial weight matrix.

    Spatial weight matrix (W). A spatial weight matrix defines the spatial arrangement of observations across space. The elements of the matrix, wi,j, express the presence or absence and the degree of spatial interaction between each possible pair of locations. We use a distance-based weight matrix. Hence, wi,j is computed as the inverse of the squared distance between geographic centroids of each pair of census divisions (CDs). Our centroid measure is the geographic centre of each CD. We assume that interaction exists within the distance radius of 1,200 km between CD centroids and beyond this radius the value of the weight is set equal to zero. The matrix is row standardized. The same spatial weight matrix is used in all the computations reported below.

    Spatial autocorrelation of value chain segment. Spatial autocorrelation occurs when the spatial distribution of the variable of interest exhibits a systematic pattern. The degree of spatial autocorrelation of segments of the value chain is measured by the Moran's I statistic, which is defined for each reference year as follows:

    Equation 1

    Description for Equation 1

    where wij is the element of the spatial weight matrix W for the observation pair i,j; x is the indicator of concern for the locations i and j, with mean equal to μ; n is the number of observations and s is a scaling factor equal to the sum of elements of W, which in the specific case of raw-standardized matrix is also equal to n (since each row sums to 1). Moran's I is similar but not equivalent to a correlation coefficient. However, the statistic is not centered around 0. The theoretical mean of Moran's I is in fact -1/n-1. But the value of Moran's I ranges from –1 (perfect negative spatial autocorrelation) to +1 (perfect positive spatial autocorrelation).

    Assessing spatial association between segments of the value chain. It can be shown that the Moran's I is equivalent to the slope of the regression line between the local indicator and its spatial lag. The "local indicator" that we are using in this paper is the "location quotient", as defined in Box 4. To assess the degree of spatial correlation between a segment of a value chain and another segment of the same value chain, we estimate a set of regressions:

    Equation 2

    Description for Equation 2
    where WLQ is the spatial lag value of the location quotient for a segment i of the value chain a, while LQj is the location quotient for the segment j (i≠j) of the same value chain a. In the table, we report the value of the estimated β coefficients.

    Mapping of reliant regions and regional milieu. The data used in the maps is the classification generated by a Moran scatterplot. This is a plot of Wz versus z, where W denotes a row-standardized spatial weights matrix and z is the standardized variable, which in our case is the location quotient of the region i for the value chain a and component b. Each map classifies the regions into four categories:
    a) regions with an above average regional intensity of employment (i.e. a LQ>1) and an above average intensity of employment in the milieu surrounding the given region (i.e. a relatively high spatial autocorrelation). This group is the core cluster of high reliant regions;
    b) regions above average on the former and below average on the latter;
    c) regions below average on the former and above average on the latter; and
    d) regions below average for both indicators. This group is the cluster of non-reliant regions. 

    The spatial measures of autocorrelation are computed using Stata modules for spatial analysis (Pisati 2001).

    End of text box 3

    Box 4
    Methodology: concentration and location

    Locational Gini coefficient. The Gini coefficient is one of the most used measures of concentration. We apply the locational version as developed by Krugman (1993). To compute the locational Gini coefficient (LG) for industry e, we calculate the region i share of employment in that industry (Se,i), relative to the national employment in that industry and relative to the regional share of employment of all industries (Stot,i). The locational Gini coefficient is then computed using the following specification:

    Equation 3

    Description for Equation 3

    where Equation 3 continued and re,i= Se,i/Stot,i with re,1< re,2< … < re,n.
    Hence, to compute the cumulative shares, the ranking of the regions is done according to the ratio re,i = Se,i/Stot,i where the region with the lowest ratio is assigned the rank number one. The value of this coefficient ranges from 1 to 0, with value 1 implying maximum concentration in one single location and 0 implying a perfectly even distribution across regions.

    Location Quotient. The location quotient (LQ) provides a measure of the intensity of employment in a given sector in a region relative to the level of employment in that sector at the national level. The LQ is presented as the ratio of the percent of the total regional employment in a sector to the percent of the total employment in that sector at the national level. A LQ is calculated for each component of the value chain. For example:

    where LQe,i is the relative intensity of employment in sector "e" in region "i" and LFe,i is the experienced labour force in sector "e" in region "i".

    Equation 4

    Description for Equation 4
    The critical values of the LQ are as follows: LQ >1 indicates that the region has a higher intensity of employment relative to the nation. LQ = 1 indicates that the region has the same intensity of employment relative to the nation. LQ < 1 indicates that the region has a lower intensity of employment relative to the national level.

    Thus, a region with an LQ>1 is relatively "specialized" or relatively "intensive", relative to the national average.

    End of text box 4

    Report a problem on this page

    Is something not working? Is there information outdated? Can't find what you're looking for?

    Please contact us and let us know how we can help you.

    Privacy notice

    Date modified: