Analytical Studies Branch Research Paper Series
The Impact of the Manufacturing Decline on Local Labour Markets in Canada

Abstract

This study quantifies the impact of the manufacturing decline on the wages and employment rates of Canadian workers in their local labour markets. The estimates, drawn from census data from 2000 to 2015, indicate that the decline in manufacturing employment had a sizable adverse effect on the wages and full-year, full-time employment rates of men—especially less educated men. In contrast, relatively few groups of women appear to have been negatively affected by the decline in manufacturing employment. The results also suggest that two-thirds or more of the decline in full-year, full-time employment rates among men observed from 2000 to 2015 in census metropolitan areas such as Montréal, Ottawa–Gatineau, Windsor, Oshawa, Toronto, Hamilton, St. Catharines–Niagara, Kitchener–Cambridge–Waterloo and Guelph can be attributed to the manufacturing decline.

Keywords: manufacturing, automation, local labour markets, wages, employment

Executive summary

From the early 2000s to the mid-2010s, the number of employees in manufacturing fell by roughly half a million in Canada. During that period, the percentage of Canadian men aged 21 to 55 employed mainly full time for at least 48 weeks in a given year fell by 5 percentage points, from 63.6% in 2000 to 58.6% in 2015. This study investigates whether the two trends are connected, i.e., whether the decline in manufacturing employment caused a decline in employment rates and wages among men.

This question is important for a variety of reasons. First, manufacturing used to be a major source of employment for less educated men. The disappearance of manufacturing jobs might therefore decrease employment opportunities for these workers. Second, many manufacturing jobs paid higher-than-average wages. As these jobs disappear, the external options for less educated individuals fall for both those who formerly held jobs in manufacturing and others. This reduces the individual bargaining power of these workers when negotiating wages. Third, through input-output linkages, the manufacturing decline may reduce labour demand in other industries, putting additional downward pressure on the wages of some workers in local labour markets. For these reasons, the manufacturing decline might reduce the wages and employment rates of less educated individuals.

Using census data from 2000 to 2015, the study finds that, on average, a 5 percentage point decline in the share of the population employed in manufacturing in a given census metropolitan area (CMA) or census agglomeration led to a 4.5 percentage point decline in full-year, full-time employment rates among men and at least a 6.9% decline in their real weekly wages. Estimated wage effects are more substantial for less educated men than for men with a bachelor’s degree or higher education. In contrast, the results indicate that relatively few groups of women appear to have been adversely affected by the decline in manufacturing employment.

The results also suggest that two-thirds or more of the decline in full-year, full-time employment rates among men observed from 2000 to 2015 in CMAs such as Montréal, Ottawa–Gatineau, Windsor, Oshawa, Toronto, Hamilton, St. Catharines–Niagara, Kitchener–Cambridge–Waterloo and Guelph can be attributed to the manufacturing decline.

1 Introduction

From the early 2000s to the mid-2010s, the number of employees in manufacturing fell by roughly half a million in Canada. During that period, the percentage of Canadian men aged 21 to 55 employed mainly full time for at least 48 weeks in a given year fell by 5 percentage points, from 63.6% in 2000 to 58.6% in 2015. This study investigates whether the two trends are connected, i.e., whether the decline in manufacturing employment caused a decline in employment rates and wages among men.

This question is important for a variety of reasons. Manufacturing used to be a major source of employment for less educated men. The disappearance of manufacturing jobs might therefore decrease employment opportunities for these workers. In addition, many manufacturing jobs paid higher-than-average wages. As these jobs disappear, the external options for less educated individuals fall for both those who formerly held jobs in manufacturing and others. This thereby reduces the individual bargaining power of these workers when negotiating wages. Finally, through input-output linkages, the manufacturing decline may reduce labour demand in other sectors, thereby putting additional downward pressure on the wages of some workers in local labour markets. For these reasons, the manufacturing decline might reduce the wages and employment rates of less educated individuals.

An alternative view is that labour markets are fairly flexible, and the initial declines in the employment and wages of affected workers will prompt them to move to more dynamic local labour markets. This view argues that while the disappearance of manufacturing jobs is expected to reduce wages and employment rates initially, labour market adjustments through migration should fully offset the initial wage and employment rate declines in the longer term.

This view can be challenged. If labour mobility is imperfect for a variety of reasons, labour supply movements might not fully offset the initial wage declines. In addition, even if employment rates eventually return to their initial levels, changes away from high-paying jobs in the industrial composition of employment might reduce the external options of workers, thereby reducing their individual bargaining power and, through this channel, reducing wages in the affected areas (Beaudry, Green and Sand 2012; Green et al. 2019).

It is therefore unclear whether the decline in manufacturing employment will adversely affect workers’ labour market outcomes. The goal of this study is to shed light on this question.^Note

The study does not attempt to identify which factors caused the decline in manufacturing employment. Labour-saving technological changes and import competition, among other factors, have been cited as potential explanations (Mowat Centre 2014). Yet identifying whether declines in local labour demand are driven by technological changes or by growth in international trade is a challenging task because the two phenomena are interrelated and because of the need to identify the impact of numerous technologies and changes in tariff and non-tariff barriers (Fort, Pierce and Schott 2018). Since the manufacturing decline is arguably driven at least by both of these factors, it provides a useful natural experiment of the degree to which these two external forces, taken together, have affected Canadian local labour markets in recent years by reducing labour demand in a specific sector: manufacturing.

2 Data and methods

The study mainly uses data from the Census of Population for 2001 and 2016. The sample consists of individuals who were aged 21 to 55 and lived in census metropolitan areas (CMAs) and census agglomerations (CAs) during the census reference week.

The study analyzes five labour market outcomes:

1. changes in employment rates from May or June 2001 to May or June 2016, where employment rates measure the percentage of the population employed during the census reference week
2. changes in average weeks worked (including zero) by individuals from 2000 to 2015
3. changes in full-year, full-time employment rates of individuals from 2000 to 2015, where full-year, full-time employment rates measure the percentage of the population that was employed mainly full time and worked at least 48 weeks in 2000 or 2015
4. changes in log average real weekly wages received by paid workers from 2000 to 2015^Note (these average weekly wages are adjusted to control for changes in the composition of the paid worker population by age, education, gender and full-time status from 2000 to 2015^Note )
5. changes in the log population of various groups of individuals from May or June 2001 to May or June 2016.

The study quantifies the degree to which the manufacturing decline observed from 2000 to 2015 in various CMAs and CAs reduced the employment rates; weeks worked; full-year, full-time employment rates; wages; and population of the individuals residing in these CMAs and CAs. To do so, the following equation is estimated:

$$\Delta Y_{rt}^g=\alpha +\beta 1*\Delta MSHAR{E}_{rt}+\beta 2*\Delta MSHAR{E}_{rt-1}+\Delta Z_{rt}{\lambda }_1+\text{}\text{}{X}_{r_{2001}}{\lambda }_2+\text{}{u}_{rt}\text{}\text{}\text{}(1)$$

where $\Delta Y_{rt}^g$ denotes the change in a given outcome observed for a given group g in area r from 2000 to 2015 (or 2001 to 2016 when changes in employment rates and in log population are analyzed). The key regressor is $\Delta MSHAR{E}_{rt}$, which denotes the change in the share of the population aged 21 to 55 employed in manufacturing from 2000 to 2015 in area r.

From 2000 to 2015, the Canadian labour market experienced two major shocks: the housing boom of the 2000s and the oil boom that took place mainly from 2001 to 2008. The vector $\Delta Z_{rt}$ controls for these two shocks. It includes changes from 2000 to 2015 in the share of the population of area r employed in (1) construction, and (2) oil and gas extraction as well as support activities for mining, and oil and gas extraction. These two variables (herein referred to as change in the share employed in construction and change in the share employed in oil) control for movements in the demand for construction workers and for workers in the natural resource sector that resulted from increases in the demand for housing and energy products.

The vector $X_{r_{2001}}$ includes several potential local labour supply shifters: (1) the share of immigrants in area r in 2001, (2) the share of individuals with a bachelor’s degree or higher education in area r in 2001, and (3) the participation rate among women in area r in 2001.^Note These variables account for the possibility that (1) CMAs and CAs with a large share of immigrants in 2001 may have attracted a relatively large number of new immigrants from 2001 to 2016, (2) CMAs and CAs with a large share of highly educated workers may have attracted proportionately more new workers during that period, and (3) CMAs and CAs with relatively low participation rates among women in 2001 may have experienced larger increases in the labour supply of women during that period.

If the local labour market adjustment to shocks is gradual, contemporaneous changes in labour market outcomes might partly capture the effect of past changes in local labour demand (Jaeger, Ruist and Stuhler 2018). For example, contemporaneous declines in wages might partly reflect previous declines in the relative importance of manufacturing. To control for this possibility, Equation (1) includes $\Delta MSHAR{E}_{rt-1}$, which is the change in the share of the population aged 21 to 55 employed in manufacturing from 1985 to 2000 in area r.^{Note Note Note}

As Charles, Hurst and Schwartz (2018) point out, there are at least two threats to the identification of β1. First, an increase in local labour demand outside manufacturing could pull individuals out of manufacturing while simultaneously increasing local employment rates. Second, local declines in labour supply can simultaneously reduce local employment rates while drawing individuals out of the manufacturing sector. To overcome these endogeneity issues, the share of the population of area r employed in manufacturing in 2000 ($MSHAR{E}_{r2000}$) is used as an instrumental variable for the key regressor ($\Delta MSHAR{E}_{rt}$). The rationale for this instrumental variable is that areas that relied heavily on manufacturing in 2000 were more exposed than other areas to the labour-saving shocks of automation and import competition from 2000 to 2015. As a result, the manufacturing share is expected to fall more in these areas than in other areas. Chart 1 confirms this hypothesis: areas that had relatively high shares of the population employed in manufacturing in 2000 generally experienced greater declines in their shares employed in manufacturing from 2000 to 2015.

Data table for Chart 1



Data table for Chart 1
Table summary
This table displays the results of Data table for Chart 1. The information is grouped by Census metropolitan area or census agglomeration (appearing as row headers), Manufacturing share in 2000, Y and Predicted values, calculated using percent and percentage points units of measure (appearing as column headers).

Census metropolitan area or census agglomeration	Manufacturing share in 2000	Y	Predicted values
	percent	percentage points
St. John's	3.9	-0.3	-1.05
Bay Roberts	7.9	-3.0	-2.49
Grand Falls-Windsor	8.5	-6.3	-2.67
Corner Brook	9.2	-4.4	-2.91
Charlottetown	5.6	-1.3	-1.65
Summerside	13.6	0.9	-4.49
Halifax	4.7	-0.8	-1.35
Kentville	12.8	-3.5	-4.19
Truro	13.1	-4.9	-4.31
New Glasgow	17.1	-9.5	-5.70
Cape Breton	4.5	-1.3	-1.29
Moncton	8.0	-3.0	-2.50
Saint John	7.8	-2.1	-2.45
Fredericton	3.7	-1.1	-0.98
Bathurst	8.5	-2.0	-2.68
Miramichi	12.3	-7.0	-4.03
Campbellton	7.6	-1.2	-2.38
Edmundston	17.5	-6.4	-5.87
Matane	14.3	-2.5	-4.73
Rimouski	4.3	0.4	-1.19
Rivière-du-Loup	12.9	0.7	-4.24
Baie-Comeau	20.7	-8.6	-6.97
Saguenay	13.2	-3.6	-4.33
Alma	13.5	-1.4	-4.44
Dolbeau-Mistassini	13.7	-3.9	-4.53
Sept-Îles	7.6	0.4	-2.38
Québec	8.2	-1.3	-2.59
Saint-Georges	24.7	-5.3	-8.39
Thetford Mines	18.8	-0.5	-6.32
Sherbrooke	19.1	-6.7	-6.43
Cowansville	27.2	-7.5	-9.27
Victoriaville	21.5	-4.9	-7.28
Trois-Rivières	14.8	-3.7	-4.91
Shawinigan	18.6	-6.1	-6.23
Drummondville	25.9	-7.1	-8.81
Granby	31.0	-10.0	-10.63
Saint-Hyacinthe	20.8	-4.6	-7.03
Sorel-Tracy	24.6	-7.8	-8.37
Joliette	12.9	-3.2	-4.25
Saint-Jean-sur-Richelieu	18.2	-5.9	-6.11
Montréal	14.5	-6.1	-4.79
Salaberry-de-Valleyfield	24.1	-12.5	-8.20
Lachute	17.7	-5.0	-5.93
Val-d'Or	7.2	-1.7	-2.23
Amos	8.6	-2.2	-2.71
Rouyn-Noranda	6.6	-1.7	-2.02
Cornwall	19.9	-11.8	-6.69
Hawkesbury	23.5	-9.9	-7.99
Ottawa–Gatineau	7.2	-4.5	-2.23
Brockville	21.0	-11.2	-7.11
Pembroke	9.1	-5.6	-2.90
Petawawa	3.4	-2.3	-0.90
Kingston	6.9	-3.1	-2.10
Belleville	16.0	-5.8	-5.32
Cobourg	21.2	-9.3	-7.16
Port Hope	22.9	-12.0	-7.77
Peterborough	11.8	-5.5	-3.85
Kawartha Lakes	13.3	-6.4	-4.37
Centre Wellington	23.5	-8.5	-7.96
Oshawa	17.6	-9.8	-5.90
Ingersoll	31.1	-7.7	-10.67
Toronto	13.8	-6.8	-4.56
Hamilton	17.2	-7.3	-5.75
St. Catharines–Niagara	16.0	-8.3	-5.32
Kitchener–Cambridge–Waterloo	24.1	-9.9	-8.20
Brantford	24.1	-8.4	-8.19
Woodstock	27.2	-4.4	-9.27
Tillsonburg	29.0	-5.9	-9.93
Norfolk	19.1	-4.4	-6.43
Guelph	23.8	-6.9	-8.09
Stratford	29.5	-9.3	-10.09
London	14.7	-5.2	-4.87
Chatham-Kent	22.4	-10.9	-7.60
Leamington	22.4	-6.9	-7.57
Windsor	25.9	-9.2	-8.82
Sarnia	15.0	-5.3	-4.98
Owen Sound	14.4	-5.2	-4.78
Collingwood	19.3	-12.9	-6.49
Barrie	15.6	-6.6	-5.20
Orillia	9.8	-3.5	-3.16
Midland	24.3	-11.7	-8.25
North Bay	6.3	-2.3	-1.90
Greater Sudbury	5.7	-1.7	-1.69
Elliot Lake	3.0	-1.3	-0.73
Temiskaming Shores	7.5	-0.6	-2.35
Timmins	5.1	-2.3	-1.49
Sault Ste. Marie	13.1	-4.9	-4.31
Thunder Bay	10.0	-5.4	-3.21
Kenora	8.6	-4.9	-2.73
Winnipeg	12.2	-4.8	-4.01
Steinbach	16.0	-2.4	-5.33
Portage la Prairie	8.5	-0.4	-2.69
Brandon	9.9	0.6	-3.18
Thompson	3.1	0.8	-0.79
Regina	4.9	-0.4	-1.41
Yorkton	6.7	1.4	-2.04
Moose Jaw	7.4	-2.2	-2.30
Swift Current	6.8	-2.7	-2.10
Saskatoon	7.8	-2.9	-2.43
North Battleford	6.9	-4.6	-2.12
Prince Albert	5.7	-4.3	-1.71
Estevan	3.4	0.7	-0.87
Medicine Hat	8.5	-4.6	-2.69
Brooks	15.5	-0.3	-5.15
Lethbridge	9.3	-2.1	-2.96
Okotoks	6.5	-1.8	-1.98
High River	13.2	-1.9	-4.33
Calgary	8.3	-3.8	-2.61
Strathmore	11.1	-5.3	-3.59
Canmore	4.2	-0.9	-1.18
Red Deer	8.0	-0.9	-2.51
Sylvan Lake	6.7	-1.7	-2.06
Lacombe	8.5	-2.1	-2.69
Camrose	5.9	-0.1	-1.78
Edmonton	8.0	-2.5	-2.50
Lloydminster	7.4	-3.7	-2.31
Cold Lake	1.3	0.7	-0.15
Grande Prairie	5.9	-1.7	-1.75
Wood Buffalo	1.8	1.1	-0.31
Wetaskiwin	7.2	-1.5	-2.21
Cranbrook	7.4	-2.9	-2.28
Penticton	8.9	-2.6	-2.83
Kelowna	9.5	-4.7	-3.03
Vernon	10.1	-3.1	-3.26
Salmon Arm	10.5	-2.4	-3.38
Kamloops	6.3	-1.7	-1.90
Chilliwack	8.0	-0.7	-2.52
Abbotsford–Mission	11.4	-2.8	-3.69
Vancouver	8.0	-2.9	-2.50
Squamish	6.6	-3.8	-2.03
Victoria	3.8	-0.8	-1.02
Duncan	11.9	-5.2	-3.89
Nanaimo	6.3	-2.6	-1.92
Parksville	5.9	-2.3	-1.76
Port Alberni	16.6	-9.6	-5.54
Courtenay	4.2	-1.6	-1.16
Campbell River	9.4	-5.8	-3.00
Powell River	13.5	-6.7	-4.44
Williams Lake	13.6	-4.6	-4.47
Quesnel	18.2	-1.1	-6.11
Prince Rupert	15.3	-10.4	-5.10
Terrace	7.5	-3.6	-2.34
Prince George	10.8	-3.8	-3.50
Dawson Creek	6.9	-2.6	-2.13
Fort St. John	3.1	1.4	-0.79
Note: The boundaries used to define census metropolitan areas and census agglomerations are those of 2011. Source: Statistics Canada, Census of Population, 2001 and 2016.

Equation (1) is estimated for various g groups: men and women of various ages (21 to 35 years and 36 to 55 years) and education levels (high school or less, postsecondary education below a bachelor’s degree, and a bachelor’s degree or higher education). Whenever Equation (1) is estimated, the dependent variable is group-specific, but all explanatory variables remain unchanged across groups.^Note

The study gives a causal interpretation to β1 but not to λ₁. This is because the two threats to identification outlined above also apply to λ₁. Without instrumental variables for changes in the shares employed in construction and in oil, λ₁ cannot be given a causal interpretation. Nevertheless, β1 can be given a causal interpretation as long as the instrumental variable, $MSHAR{E}_{r2000}$, is uncorrelated with the error term U_rt after conditioning results on the set of controls included in Equation (1) (Stock and Watson 2011). This hypothesis is the identifying assumption of the study.

Equation (1) captures the direct effect of changes in the share of the population employed in manufacturing on labour market outcomes. That is, it measures this effect while holding constant changes in the share employed in construction and the share employed in oil. Since changes in manufacturing employment affect economic activity, they will likely also affect the housing market (through an income effect that will affect housing demand) and the share employed in oil (through movements in the demand for energy from manufacturing firms). These indirect effects, which induce movements in the demand for construction workers and workers in the oil and gas sectors, are not captured by Equation (1), but reflect part of the total impact of changes in manufacturing employment. Since these indirect effects are expected to be positive, β1 will provide a conservative estimate of the overall impact that the manufacturing decline will have on local labour markets.^Note

One concern with estimating Equation (1) for all 145 CMAs and CAs of the 10 Canadian provinces is that those located in the oil-producing provinces of Newfoundland and Labrador, Alberta, and Saskatchewan faced a more dynamic economic environment than others from 2000 to 2015. As a result, the nationwide average effects might conceal important cross-regional differences. To address this concern, Equation (1) is also estimated for the subset of 115 CMAs and CAs that are located in the seven non-oil-producing provinces (Prince Edward Island, Nova Scotia, New Brunswick, Quebec, Ontario, Manitoba and British Columbia).

Throughout the paper, heteroscedasticity-robust standard errors are used.

3 Descriptive evidence

From 2001 to 2016, the number of Canadian employees working in manufacturing fell by roughly half a million, dropping from 1.98 million in 2001 to 1.48 million in 2016 (Table 1). Six industries accounted for about 60% of the employment decline: clothing manufacturing (12.8%), transportation equipment manufacturing (10.3%), computer and electronic product manufacturing (10.2%), paper manufacturing (10.2%), wood product manufacturing (8.8%), and primary metal manufacturing (7.5%).

The magnitude of the decline in the relative importance of manufacturing differed across CMAs and CAs. In Ontario, the share of the population aged 21 to 55 employed in manufacturing fell by 6.8 percentage points or more in Toronto, Hamilton, St. Catharines–Niagara, Kitchener–Cambridge–Waterloo, Brantford, Guelph, Stratford and Windsor (Table 2). In contrast, the share employed in manufacturing fell by no more than 3.8 percentage points in several CMAs and CAs in the Atlantic provinces, Saskatchewan, Alberta and British Columbia.^Note

The data shown in Charts 2 and 3 suggest that the manufacturing decline caused a decline in the employment rates and wages of men. Areas that experienced a relatively large decline in the share employed in manufacturing generally displayed less favourable movements in full-year, full-time employment rates (Chart 2) and wages (Chart 3) among men than other areas. The next section investigates whether these patterns hold in multivariate analyses for men and women of different ages and education levels.

Data table for Chart 2



Data table for Chart 2
Table summary
This table displays the results of Data table for Chart 2. The information is grouped by Census metropolitan area or census agglomeration (appearing as row headers), Change in the manufacturing share, Y and Predicted values, calculated using percentage points units of measure (appearing as column headers).

Census metropolitan area or census agglomeration	Change in the manufacturing share	Changes in the percentage of men employed full year, full time (percentage points)	Predicted values
	percentage points
St. John's	-0.3	2.6	-2.2
Bay Roberts	-3.0	13.5	-3.6
Grand Falls-Windsor	-6.3	-7.0	-5.4
Corner Brook	-4.4	2.9	-4.4
Charlottetown	-1.3	-1.9	-2.7
Summerside	0.9	-2.5	-1.5
Halifax	-0.8	-3.6	-2.4
Kentville	-3.5	1.2	-3.9
Truro	-4.9	-5.0	-4.6
New Glasgow	-9.5	-6.1	-7.2
Cape Breton	-1.3	1.8	-2.7
Moncton	-3.0	0.2	-3.6
Saint John	-2.1	0.5	-3.1
Fredericton	-1.1	-3.1	-2.6
Bathurst	-2.0	-4.2	-3.1
Miramichi	-7.0	-3.7	-5.8
Campbellton	-1.2	3.4	-2.6
Edmundston	-6.4	-0.1	-5.5
Matane	-2.5	12.0	-3.3
Rimouski	0.4	2.0	-1.8
Rivière-du-Loup	0.7	-0.3	-1.6
Baie-Comeau	-8.6	-3.9	-6.7
Saguenay	-3.6	-2.3	-3.9
Alma	-1.4	-3.8	-2.7
Dolbeau-Mistassini	-3.9	-0.7	-4.1
Sept-Îles	0.4	-0.8	-1.8
Québec	-1.3	-0.7	-2.7
Saint-Georges	-5.3	-3.5	-4.9
Thetford Mines	-0.5	0.5	-2.2
Sherbrooke	-6.7	-6.0	-5.6
Cowansville	-7.5	2.4	-6.1
Victoriaville	-4.9	-6.0	-4.7
Trois-Rivières	-3.7	-2.9	-4.0
Shawinigan	-6.1	-4.5	-5.3
Drummondville	-7.1	-2.6	-5.8
Granby	-10.0	-9.0	-7.4
Saint-Hyacinthe	-4.6	-7.8	-4.5
Sorel-Tracy	-7.8	-3.2	-6.3
Joliette	-3.2	-9.3	-3.7
Saint-Jean-sur-Richelieu	-5.9	-4.0	-5.2
Montréal	-6.1	-6.8	-5.3
Salaberry-de-Valleyfield	-12.5	-11.8	-8.8
Lachute	-5.0	-8.3	-4.7
Val-d'Or	-1.7	1.8	-2.9
Amos	-2.2	8.3	-3.2
Rouyn-Noranda	-1.7	-0.6	-2.9
Cornwall	-11.8	-9.3	-8.4
Hawkesbury	-9.9	-13.1	-7.4
Ottawa–Gatineau	-4.5	-6.2	-4.4
Brockville	-11.2	-9.8	-8.1
Pembroke	-5.6	-5.9	-5.0
Petawawa	-2.3	-4.3	-3.2
Kingston	-3.1	-1.0	-3.7
Belleville	-5.8	-5.2	-5.2
Cobourg	-9.3	-11.4	-7.1
Port Hope	-12.0	-8.7	-8.5
Peterborough	-5.5	-8.8	-5.0
Kawartha Lakes	-6.4	-9.5	-5.5
Centre Wellington	-8.5	-5.1	-6.6
Oshawa	-9.8	-11.1	-7.3
Ingersoll	-7.7	-4.0	-6.2
Toronto	-6.8	-8.8	-5.7
Hamilton	-7.3	-7.7	-6.0
St. Catharines–Niagara	-8.3	-9.5	-6.5
Kitchener–Cambridge–Waterloo	-9.9	-9.9	-7.4
Brantford	-8.4	-6.7	-6.6
Woodstock	-4.4	-3.9	-4.4
Tillsonburg	-5.9	-5.0	-5.2
Norfolk	-4.4	-7.4	-4.4
Guelph	-6.9	-6.6	-5.8
Stratford	-9.3	-6.1	-7.1
London	-5.2	-8.5	-4.8
Chatham-Kent	-10.9	-8.9	-7.9
Leamington	-6.9	0.3	-5.7
Windsor	-9.2	-12.7	-7.0
Sarnia	-5.3	-10.4	-4.9
Owen Sound	-5.2	-6.2	-4.8
Collingwood	-12.9	-9.0	-9.0
Barrie	-6.6	-9.1	-5.6
Orillia	-3.5	-10.6	-3.9
Midland	-11.7	-7.9	-8.3
North Bay	-2.3	-6.4	-3.2
Greater Sudbury	-1.7	2.9	-2.9
Elliot Lake	-1.3	-6.6	-2.7
Temiskaming Shores	-0.6	-4.7	-2.3
Timmins	-2.3	-1.6	-3.2
Sault Ste. Marie	-4.9	-5.0	-4.6
Thunder Bay	-5.4	-4.6	-4.9
Kenora	-4.9	-1.5	-4.6
Winnipeg	-4.8	-7.1	-4.6
Steinbach	-2.4	-5.9	-3.3
Portage la Prairie	-0.4	-0.2	-2.2
Brandon	0.6	-5.5	-1.7
Thompson	0.8	-7.5	-1.5
Regina	-0.4	-2.6	-2.2
Yorkton	1.4	-1.5	-1.2
Moose Jaw	-2.2	-0.1	-3.2
Swift Current	-2.7	-2.5	-3.5
Saskatoon	-2.9	-3.1	-3.6
North Battleford	-4.6	-4.5	-4.5
Prince Albert	-4.3	-3.0	-4.3
Estevan	0.7	-9.9	-1.6
Medicine Hat	-4.6	-8.9	-4.5
Brooks	-0.3	-7.4	-2.1
Lethbridge	-2.1	-5.8	-3.1
Okotoks	-1.8	-9.7	-3.0
High River	-1.9	-4.3	-3.0
Calgary	-3.8	-8.0	-4.1
Strathmore	-5.3	-10.8	-4.9
Canmore	-0.9	-8.7	-2.5
Red Deer	-0.9	-8.2	-2.5
Sylvan Lake	-1.7	-12.7	-2.9
Lacombe	-2.1	-8.4	-3.2
Camrose	-0.1	-7.6	-2.0
Edmonton	-2.5	-6.3	-3.4
Lloydminster	-3.7	-9.3	-4.0
Cold Lake	0.7	-17.0	-1.6
Grande Prairie	-1.7	-7.0	-2.9
Wood Buffalo	1.1	-0.2	-1.4
Wetaskiwin	-1.5	-2.4	-2.8
Cranbrook	-2.9	-5.1	-3.6
Penticton	-2.6	-2.2	-3.4
Kelowna	-4.7	-2.0	-4.6
Vernon	-3.1	-3.1	-3.7
Salmon Arm	-2.4	2.0	-3.3
Kamloops	-1.7	-1.2	-2.9
Chilliwack	-0.7	0.5	-2.4
Abbotsford–Mission	-2.8	-2.8	-3.5
Vancouver	-2.9	-2.7	-3.6
Squamish	-3.8	4.0	-4.1
Victoria	-0.8	-2.4	-2.4
Duncan	-5.2	-1.9	-4.8
Nanaimo	-2.6	-0.3	-3.4
Parksville	-2.3	0.3	-3.3
Port Alberni	-9.6	-6.4	-7.2
Courtenay	-1.6	0.2	-2.9
Campbell River	-5.8	-4.6	-5.1
Powell River	-6.7	-7.0	-5.6
Williams Lake	-4.6	0.5	-4.5
Quesnel	-1.1	-1.7	-2.6
Prince Rupert	-10.4	-0.1	-7.7
Terrace	-3.6	-2.5	-3.9
Prince George	-3.8	-1.9	-4.0
Dawson Creek	-2.6	-4.3	-3.4
Fort St. John	1.4	0.5	-1.2
Note: The boundaries used to define census metropolitan areas and census agglomerations are those of 2011. Source: Statistics Canada, Census of Population, 2001 and 2016.

Data table for Chart 3



Data table for Chart 3
Table summary
This table displays the results of Data table for Chart 3. The information is grouped by Census metropolitan area or census agglomeration (appearing as row headers), Change in the manufacturing share, Y and Predicted values, calculated using percentage points and logarithmic value units of measure (appearing as column headers).

Census metropolitan area or census agglomeration	Change in the manufacturing share	Changes in the log real weekly wages of men	Predicted values
	percentage points	logarithmic value
St. John's	-0.3	0.32	0.22
Bay Roberts	-3.0	0.42	0.16
Grand Falls-Windsor	-6.3	0.30	0.09
Corner Brook	-4.4	0.23	0.13
Charlottetown	-1.3	0.07	0.20
Summerside	0.9	0.02	0.24
Halifax	-0.8	0.08	0.21
Kentville	-3.5	0.12	0.15
Truro	-4.9	0.12	0.12
New Glasgow	-9.5	0.10	0.02
Cape Breton	-1.3	0.25	0.19
Moncton	-3.0	0.06	0.16
Saint John	-2.1	0.10	0.18
Fredericton	-1.1	0.11	0.20
Bathurst	-2.0	0.10	0.18
Miramichi	-7.0	0.12	0.07
Campbellton	-1.2	0.10	0.20
Edmundston	-6.4	0.01	0.08
Matane	-2.5	0.10	0.17
Rimouski	0.4	0.10	0.23
Rivière-du-Loup	0.7	0.10	0.24
Baie-Comeau	-8.6	0.05	0.04
Saguenay	-3.6	0.04	0.15
Alma	-1.4	0.05	0.19
Dolbeau-Mistassini	-3.9	-0.03	0.14
Sept-Îles	0.4	0.19	0.23
Québec	-1.3	0.06	0.19
Saint-Georges	-5.3	0.06	0.11
Thetford Mines	-0.5	0.00	0.21
Sherbrooke	-6.7	0.04	0.08
Cowansville	-7.5	-0.01	0.06
Victoriaville	-4.9	0.05	0.12
Trois-Rivières	-3.7	0.02	0.14
Shawinigan	-6.1	-0.02	0.09
Drummondville	-7.1	0.05	0.07
Granby	-10.0	0.06	0.01
Saint-Hyacinthe	-4.6	0.05	0.12
Sorel-Tracy	-7.8	0.03	0.05
Joliette	-3.2	0.02	0.15
Saint-Jean-sur-Richelieu	-5.9	0.08	0.10
Montréal	-6.1	0.01	0.09
Salaberry-de-Valleyfield	-12.5	-0.02	-0.05
Lachute	-5.0	-0.03	0.12
Val-d'Or	-1.7	0.32	0.19
Amos	-2.2	0.20	0.17
Rouyn-Noranda	-1.7	0.22	0.19
Cornwall	-11.8	0.01	-0.03
Hawkesbury	-9.9	-0.04	0.01
Ottawa–Gatineau	-4.5	0.00	0.13
Brockville	-11.2	0.07	-0.02
Pembroke	-5.6	0.14	0.10
Petawawa	-2.3	0.21	0.17
Kingston	-3.1	0.06	0.16
Belleville	-5.8	0.01	0.10
Cobourg	-9.3	0.08	0.02
Port Hope	-12.0	0.09	-0.04
Peterborough	-5.5	0.04	0.10
Kawartha Lakes	-6.4	0.07	0.09
Centre Wellington	-8.5	0.01	0.04
Oshawa	-9.8	0.00	0.01
Ingersoll	-7.7	0.05	0.06
Toronto	-6.8	0.00	0.08
Hamilton	-7.3	0.01	0.07
St. Catharines–Niagara	-8.3	-0.02	0.04
Kitchener–Cambridge–Waterloo	-9.9	0.01	0.01
Brantford	-8.4	0.01	0.04
Woodstock	-4.4	0.08	0.13
Tillsonburg	-5.9	-0.05	0.10
Norfolk	-4.4	0.06	0.13
Guelph	-6.9	0.00	0.07
Stratford	-9.3	0.02	0.02
London	-5.2	-0.01	0.11
Chatham-Kent	-10.9	-0.10	-0.01
Leamington	-6.9	-0.04	0.08
Windsor	-9.2	-0.14	0.02
Sarnia	-5.3	0.12	0.11
Owen Sound	-5.2	0.15	0.11
Collingwood	-12.9	0.14	-0.05
Barrie	-6.6	0.01	0.08
Orillia	-3.5	0.05	0.15
Midland	-11.7	-0.04	-0.03
North Bay	-2.3	0.12	0.17
Greater Sudbury	-1.7	0.15	0.19
Elliot Lake	-1.3	0.20	0.19
Temiskaming Shores	-0.6	0.23	0.21
Timmins	-2.3	0.28	0.17
Sault Ste. Marie	-4.9	0.08	0.12
Thunder Bay	-5.4	0.06	0.11
Kenora	-4.9	0.07	0.12
Winnipeg	-4.8	0.11	0.12
Steinbach	-2.4	0.14	0.17
Portage la Prairie	-0.4	0.15	0.21
Brandon	0.6	0.21	0.23
Thompson	0.8	0.32	0.24
Regina	-0.4	0.24	0.21
Yorkton	1.4	0.37	0.25
Moose Jaw	-2.2	0.35	0.18
Swift Current	-2.7	0.37	0.16
Saskatoon	-2.9	0.29	0.16
North Battleford	-4.6	0.30	0.12
Prince Albert	-4.3	0.21	0.13
Estevan	0.7	0.33	0.24
Medicine Hat	-4.6	0.32	0.12
Brooks	-0.3	0.11	0.22
Lethbridge	-2.1	0.21	0.18
Okotoks	-1.8	0.29	0.18
High River	-1.9	0.14	0.18
Calgary	-3.8	0.21	0.14
Strathmore	-5.3	0.38	0.11
Canmore	-0.9	0.26	0.20
Red Deer	-0.9	0.26	0.20
Sylvan Lake	-1.7	0.36	0.18
Lacombe	-2.1	0.39	0.18
Camrose	-0.1	0.34	0.22
Edmonton	-2.5	0.31	0.17
Lloydminster	-3.7	0.29	0.14
Cold Lake	0.7	0.47	0.24
Grande Prairie	-1.7	0.31	0.18
Wood Buffalo	1.1	0.46	0.24
Wetaskiwin	-1.5	0.38	0.19
Cranbrook	-2.9	0.27	0.16
Penticton	-2.6	0.19	0.17
Kelowna	-4.7	0.20	0.12
Vernon	-3.1	0.19	0.15
Salmon Arm	-2.4	0.12	0.17
Kamloops	-1.7	0.17	0.19
Chilliwack	-0.7	0.15	0.21
Abbotsford–Mission	-2.8	0.09	0.16
Vancouver	-2.9	0.08	0.16
Squamish	-3.8	0.04	0.14
Victoria	-0.8	0.10	0.21
Duncan	-5.2	0.11	0.11
Nanaimo	-2.6	0.13	0.17
Parksville	-2.3	0.20	0.17
Port Alberni	-9.6	0.03	0.02
Courtenay	-1.6	0.09	0.19
Campbell River	-5.8	0.15	0.10
Powell River	-6.7	0.07	0.08
Williams Lake	-4.6	0.19	0.12
Quesnel	-1.1	0.15	0.20
Prince Rupert	-10.4	0.12	0.00
Terrace	-3.6	0.13	0.15
Prince George	-3.8	0.13	0.14
Dawson Creek	-2.6	0.27	0.17
Fort St. John	1.4	0.32	0.25
Note: Changes in weekly wages control for compositional effects by age, education and full-time status within a given census metropolitan area or census agglomeration. The boundaries used to define census metropolitan areas and census agglomerations are those of 2011. Source: Statistics Canada, Census of Population, 2001 and 2016.

4 Regression results

Table 3 shows results from Equation (1), estimated for men and women separately. Parameter estimates for β1 are shown for (a) weighted regressions (in which observations are weighted by the size of the population aged 21 to 55 in a given area in 2001) and unweighted regressions^Note (in which each area receives a weight equal to 1), and (b) all provinces and the seven non-oil-producing provinces.

Regardless of the weighting scheme and the set of provinces considered, the numbers indicate that the manufacturing decline reduced wages, average weeks worked, and full-year, full-time employment rates among men. For example, parameter estimates for full-year, full-time employment rates among men calculated from weighted regressions equal roughly 0.90, thereby suggesting that a 5 percentage point decline in the share of the population employed in manufacturing led to a 4.5 percentage point decline in the percentage of men working full-year, full-time in a given CMA or CA. Parameter estimates for wage changes among men in non-oil-producing provinces equal roughly 1.70, thereby suggesting that a 5 percentage point decline in the share of the population employed in manufacturing led to a 0.085 log point decline in wages among men (i.e., about an 8.5% decline in wages among men).^Note

In contrast, the data provide little support for the hypothesis that the manufacturing decline reduced wages, average weeks worked, or full-year, full-time employment rates among women.

Table 4 investigates the issue in greater depth for men. It provides parameter estimates for β1 for men of different ages and education levels.^Note Among men aged 21 to 55, the manufacturing decline reduced wages, average weeks worked, and full-year, full-time employment rates within each education level. This is true both for all 145 CMAs and CAs and for only those in the seven non-oil-producing provinces. However, in each case, the impact on wages is almost twice as high for men with no bachelor’s degree than for more educated men.

Table 4 also highlights interesting age differences among men with a postsecondary credential below a bachelor’s degree. For this group, the numbers indicate that the effect of the manufacturing decline on employment rates, weeks worked and full-year, full-time employment rates was at least twice as pronounced among younger workers (aged 21 to 35) than among older workers (aged 36 to 55). For instance, in the non-oil-producing provinces, a 5 percentage point decline in the manufacturing share was associated with a 7.2 percentage point decline in the full-year, full-time employment rates of younger men (0.05 times 1.45) and a 3.3 percentage point decline in the full-year, full-time employment rates of older men (0.05 times 0.66).

Cross-educational wage differences are also observed within each age group, especially among younger workers. Regardless of the provinces considered, the parameter estimate β1 for wage changes is at least twice as high for younger men with no bachelor’s degree than for their more educated counterparts.

In sum, while the data indicate that the impact of the manufacturing decline was fairly widespread among men (i.e., not limited to specific age groups or education levels), they also show that younger and less educated men were more adversely affected than other groups of male workers.

In contrast, relatively few groups of women appear to have been adversely affected by the decline in manufacturing employment. Regardless of the provinces and outcomes considered, no statistically significant effects are detected for women aged 36 to 55 (Table 5). While there is evidence that young women experienced a decline in wages and full-year, full-time employment rates, the estimated effects are smaller and generally less precisely estimated than for young men.

Tables 4 and 5 show no evidence that older men and women moved to more economically dynamic areas following declines in manufacturing employment. In contrast, there is robust evidence that younger men and women with a bachelor’s degree migrated to other regions in response to the manufacturing decline. For example, estimates excluding the oil-producing provinces indicate that a 5 percentage point decline in the manufacturing share of a given area led to a decline of 0.12 to 0.13 log points in the population of younger workers with a bachelor’s degree or higher education in that area.

5 Robustness checks and relevance

The wage estimates shown so far are based on the Consumer Price Index (CPI) for Canada and therefore neglect region-specific movements in the cost of living. If areas that experienced large declines in the share employed in manufacturing also experienced large declines in housing costs, the wage estimates shown so far may overestimate the magnitude of the declines in real weekly wages.

To address this issue, Table 6 compares the initial wage estimates—those based on the national CPI and referred to as Model 1—with estimates obtained after adding changes in log average home prices from 2001 to 2016 to Equation (1). The second set of estimates is referred to as Model 2. Controlling for home price growth somewhat reduces the wage estimates but does not alter the main findings. For example, parameter estimates for wage changes among men in non-oil-producing provinces drop from approximately 1.70 to 1.40, suggesting that a 5 percentage point decline in the share of the population employed in manufacturing led to a log point decline of at least 0.069 (about 6.9%) in real weekly wages among men.

Another concern is that some of the effects observed might be driven by selectivity based on unobservable factors. If the decline in manufacturing employment in a given area led to the out-migration of the best workers (within each education cell considered), then part of the decline in full-year, full-time employment rates and wages among men observed within education categories might reflect changes in the composition of the workforce toward lower-ability workers.

Table 7 investigates this issue for wage estimates. It compares initial weekly wage estimates for young men based on census data with panel data estimates based on the Canadian Employer–Employee Dynamics Database (CEEDD). The panel data estimates are based on the same group of men who were aged 21 to 35 in 2000 and were in the same CMA or CA in 2000 and 2015. While this group of workers is not representative of the population of young men in 2000, it allows for an analysis that controls for time-invariant, but unobserved, worker abilities.

Because CEEDD data can provide only estimates of annual wages, Table 7 also shows annual wage estimates for young men based on census data. Consistent with the fact that average weeks worked by men fell in response to the manufacturing decline, the annual wage estimates derived from census data are somewhat higher than the initial weekly wage estimates. More importantly, the annual wage estimates from panel data equal between 77% and 89% of the annual wage estimates from census data. This finding implies that most of the wage effects observed based on non-panel (census) data remain when using panel data. Therefore, selectivity based on unobservable factors does not play a major role.

To highlight the relevance of the main findings, Table 8 shows what they imply for specific CMAs and CAs in Quebec and Ontario, the two main manufacturing provinces. Multiplying the parameter estimate β1 obtained for changes in full-year, full-time employment rates among men living in the non-oil-producing provinces (0.91) by the change in an area’s share of the population employed in manufacturing yields an estimate of the changes in full-year, full-time employment rates resulting from the manufacturing decline. The results indicate that two-thirds or more of the decline in full-year, full-time employment rates among men from 2000 to 2015 in CMAs such as Montréal, Ottawa–Gatineau, Windsor, Oshawa, Toronto, Hamilton, St. Catharines–Niagara, Kitchener–Cambridge–Waterloo and Guelph can be attributed to the manufacturing decline.

Likewise, an estimate of the changes in real weekly wages resulting from the manufacturing decline can be calculated by multiplying the wage parameter estimate obtained from Model 1 for men living in non-oil-producing provinces (1.66), by the change in an area’s share of the population employed in manufacturing. This estimate indicates that the manufacturing decline tended to reduce real weekly wages among men in the aforementioned CMAs by around 10%.

Table 9 provides a sense of the degree to which the manufacturing decline reduced wages and full-year, full-time employment rates among men nationwide.^Note Changes in male wages and full-year, full-time employment rates resulting from the manufacturing decline are calculated as follows: multiply the group-specific parameter estimates from the left panel of Table 4 by the actual change in the share of individuals aged 21 to 55 employed in manufacturing at the national level. The results indicate that the manufacturing decline (a) accounted for at least half of the decline in full-year, full-time employment rates among men, and (b) tended to reduce wages more among less educated men than among those with a bachelor’s degree or higher education.

Overall, the numbers shown in Tables 8 and 9 indicate that the manufacturing decline had a substantial impact on the wages and full-year, full-time employment rates of men in several CMAs and CAs as well as nationwide.^Note

The main findings of the study are similar to those of Charles, Hurst and Schwartz (2018) for men, but differ for women. Canadian data from 2000 to 2015 suggest that a 5 percentage point decline in the share of the population employed in manufacturing led to a log point decline of at least 0.069 (about 6.9%) in real weekly wages among men in local labour markets. U.S. data from 2000 to 2016 suggest that the corresponding estimated decline in real hourly wages among men was very similar to the findings in this study, at about 6.2% (i.e., 1.23 times 0.05—see Table 2 from Charles, Hurst and Schwartz [2018]). A 5 percentage point decline in the share of the population employed in manufacturing reduced employment rates among men by about 2 percentage points in both countries.^Note Charles, Hurst and Schwartz (2018) found that the negative impact of the manufacturing decline on wages and employment rates was similar for men and women. In contrast, the Canadian data provide little evidence that the manufacturing decline reduced wages and employment rates among women. Which factors underlie this cross-country difference is a question for future research.

In line with Gould (2019), data from both countries suggest that the manufacturing decline tended to increase wage inequality between less educated men and their better-educated counterparts within local labour markets. This can be seen by noting that the wage parameter estimate for men aged 21 to 55 with high school or less education is, when considering all provinces, about twice as large as that for men with a bachelor’s degree or more (Table 6). In the United States, Charles, Hurst and Schwartz (2018) obtain a parameter estimate that is about three times as large for less educated men than for those with a bachelor’s degree.^Note

6 Conclusion

This study quantifies the impact of the manufacturing decline on the wages and employment rates of Canadian workers in their local labour markets. The results reject the view that local labour markets return to their initial (full-year, full-time) employment rates several years after a sector-specific negative labour demand shock. Instead, they suggest that movements in full-year, full-time employment rates among men have a structural component. The estimates, obtained over a 15-year period, indicate that the manufacturing decline had a sizable adverse effect on the wages and full-year, full-time employment rates of men. In contrast, relatively few groups of women appear to have been adversely affected by the decline in manufacturing employment.

7 Tables

        

Appendix

 

References

Beaudry, P., D.A. Green, and B. Sand. 2012. “Does industrial composition matter for wages? A test of search and bargaining theory.” Econometrica 80 (3): 1063–1104.

Charles, K., E. Hurst, and M. Notowidigdo. 2016. “The masking of the decline in manufacturing employment by the housing bubble.” Journal of Economic Perspectives 30 (2): 179–200.

Charles, K., E. Hurst, and M. Notowidigdo. 2019. “Housing booms, manufacturing decline and labour market outcomes.” The Economic Journal 129 (617): 209–248.

Charles, K., E. Hurst, and M. Schwartz. 2018. The Transformation of Manufacturing and the Decline in U.S. Employment. NBER Working Paper Series, no. 24468. Cambridge, Massachusetts: National Bureau of Economic Research.

Fort, T.C., J.R. Pierce, and P.K. Schott. 2018. “New perspectives on the decline of US manufacturing employment.” Journal of Economic Perspectives 32 (2): 47–72.

Gould, E.D. 2019. “Explaining the unexplained: Residual wage inequality, manufacturing decline and low-skilled immigration.” The Economic Journal 129 (619): 1281–1326.

Green, D.A., R. Morissette, B. Sand, and I. Snoddy. 2019. “Economy-wide spillovers from booms: Long distance commuting and the spread of wage effects.” Journal of Labor Economics. Forthcoming.

Jaeger, D.A., J. Ruist, and J. Stuhler. 2018. Shift-share Instruments and the Impact of Immigration. Discussion Paper Series, no. 11307. Bonn, Germany: IZA Institute of Labor Economics.

Mowat Centre. 2014. “How Ontario lost 300,000 manufacturing jobs (and why most aren’t coming back).” Mowat Centre: Ontario’s Voice on Public Policy. July 29. Available at: https://munkschool.utoronto.ca/mowatcentre.ca/how-ontario-lost-300000-manufacturing-jobs/ (accessed September 16, 2019).

Stock, J.H. and M.W. Watson. 2011. Introduction to Econometrics. Third Edition. Pearson.