Growth Firms Project: Key Findings

Attachment A

Statistics Canada's Longitudinal Business Data Bases, the Measure of Employment, and the Feasibility of Tabulations: Some Background and Discussion

Development of longitudinal business data

Statistics Canada has been engaged in developing longitudinal data bases of businesses in Canada since 1972. One such effort combines information from its Business Register (in itself an evolving and significantly improving frame of all employer businesses in Canada – and, since 1998, also non-employer businesses) with payroll data from the Canada Customs and Revenue Agency (CCRA). The original intent was to measure "job turnover – the degree to which jobs are created in newly-identified and growing businesses and lost in no-longer-identified and declining businesses."¹ The name of the endeavour, Longitudinal Employment Analysis Program, became the name of the file, LEAP. Statistics Canada's product Employment Dynamics – a year-to-year pairwise comparison of continuing, entering and exiting businesses and their associated employment – is a by-product of LEAP. The file encompasses the universe of employer businesses in Canada, that is, all entities that maintain a payroll and submit information to CCRA.

The file uses business level data obtained from Revenue Canada tax records, which do not accord well with establishment level data that are frequently used in Statistics Canada's surveys. The establishment, while being more amenable to industrial classification because it comprises a more homogeneous set of activities, has no counterpart in Revenue Canada tax data. Businesses may span one or more establishments, and include all private and public employers. Carefully tracking employment information at both the business and establishment level allows enterprise data to be reported, identified at the 3-digit SIC industrial level, by province. All analysis in this project is at the enterprise level. (See below for more on the enterprise concept and on limitations on the release of data.)

In constructing the file, considerable effort goes into identifying true continuity, or conversely, recognizing false births and false deaths of establishments, corporations and enterprises. Tests have shown that a computerized search for continuity in employment through matching of social insurance numbers is a highly effective tool. In a second round, more costly name matching exercises add further to the accuracy of the assignment of longitudinal identifiers.

The 1983-2000 version of the file has recently become available, reflecting the current state of affairs and correcting past errors. LEAP has been successfully linked with Statistics Canada's corporate tax information file and an update of this linked file to cover 1983-2001 is expected next spring.

Measuring employment

The employment measure in LEAP is derived by dividing a firm's payroll by the most appropriate average annual earnings figure available from the Survey of Employment, Payrolls and Hours (SEPH; its earnings data are, for each province, industry-and size-class specific²). The resulting measure is referred to as an Average Labour Unit or ALU. ALUs measure the number of persons, on average over the course of the year, who worked for a firm.

An expanded version of LEAP brings in data from the Small Area File (SAF), adding more detail in the geographical dimension. But LEAP/SAF also offers another measure of employment, the Individual Labour Unit or ILU. Unlike ALUs, an ILU does not require an industry average annual wage in order to be derived. Instead, the measure counts the number of T4 slips submitted by firms. T4 recipients who worked for more than one employer have their "unit" partitioned in proportion to their earnings. That makes an ILU a measure of the number of people who, over the course of a year, worked for an employer. The 1983-1999 version of LEAP/SAF was available for this project.

Either measure, ALU and ILU, is blind to part-time or seasonal work. However, provided a firm conforms to its industry average, an ALU will accurately measure what it intends to measure, even if the work is part-time or seasonal.⁴ The total quantum of ILUs in the economy (15.1 mln in 1999) is therefore larger than the total quantum of ALUs 12.8 mln).⁵ As long as one recognizes what is being measured, either measure would do with regard to its ability to account of other than full time work. An issue arises to the extent that patterns of part-time and seasonal work have changed over the course of the period considered – a consideration which needs to be further pursued.

For this study, ILUs were chosen as the measure of employment considering that growth firms may significantly deviate from industry averages – in their remuneration or with regard to the employ of part-time workers. Moreover, the growth pattern of their wages could be expected to be different. Using ILUs allows firm-specific average wage rates to be calculated and permits comparisons of average wage rates between firms and across growth groups. Under any of these conditions, an ALU measure would be biased. As noted in the main text, indications from this phase of the project are that indeed growth firms start from lower wages and increase wages faster.

Finally, one should bear in mind that neither ALU nor ILU counts owner-operators who are not on their own payroll. The full size of the SME sector is in that regard understated. Likewise, measured job creation through births excludes people not on their own payroll. The smaller the firm, the greater the likelihood that the owner-operator is not counted here. Labour market participation through contract work is likewise not accounted for here, unless the worker is on some entity's payroll.

The concept of enterprise

Another important choice made in this growth firms project was the decision to tabulate the data at the enterprise level. This is also the level at which SEPH data by size of firm are published.⁶ While establishment-level data are best suited for location-centred analysis, for growth patterns it was felt that decision making (and hence the most helpful analysis) was more likely to be found at the enterprise level.

The choice of enterprise as the level of analysis has important implications for the geographical dimension of the tabulations. In Canada-wide tabulations, establishments are aggregated to enterprises at the national level. In tabulations by province, the operation of a multi-province enterprise is counted in each province where it occurs – a provincial "enterprise" is created. There are therefore 199,000 continuing firms in the national tabulations, against 203,600 in the sum of all provincial tabulations. Likewise the classification of enterprises in types of growth firms, births or deaths is provincespecific because it is based on these "provincial enterprises." Sub-provincial data would likewise compute CMA 'enterprises' – the representation of possibly a national enterprise in the specific CMA. The regional tabulations of this project are strictly subtotals of provincial tabulations.

Disclosure limitations

Detailed tabulations by province, and even national cross-tabulations by industry and size rapidly run into confidentiality constraints. Tabulations by region, however (Atlantic, Quebec, Ontario, Prairies, B.C.) are more feasible. In both national and regional tabulations, seven size classes⁷ were distinguished and seventy 2-digit industries.

Aggregation to eight or ten subgroups of industries may allow more disclosure but a preliminary look suggests that less useful information relevant to the analysis of growth emerges at that high a level of aggregation. This will be further examined. Disaggregation to the 3-digit level would rapidly run into disclosure problems.

More disclosure could also be bought at the expense of putting the bar for hyper growth lower. Compressing size classifications – all at the high end – would provide little solace. As it stands, many suppressions (in cross-tabulations by industry) result in "20+" or "50+" aggregations; there seldom are disclosure issues for micro-firms (fewer than 5 employees).

---

¹ John Baldwin, Richard Dupuy and William Penner, "Development of Longitudinal Panel Data from Business Registers: Canadian Experience," Statistics Canada, Analytical Studies Branch Research paper No. 49, 1992, p. 3. See also Cat. No. 18-501E (1988).

² The size ranges used to calculate the ALU are: less than 19, 20 to 49, 50 to 199, and more than 200.

³ In particular, there seems to be a widespread belief that ALUs represent a full-time-equivalent. This is not in general the case. Assume a firm pays one worker $50 for a 10-hour week and another worker $150 for a 40-hour week, for a total payroll of $200. Assume the relevant average industry wage is $100/week. The resulting ALU measure is 2, while the full-time-equivalent is 1¼.

⁴ For an extreme example, assume a firm employs 12 people over the course of the year, one each for one month. Provided other firms in the industry do the same on average and pay the same wage, the ALU measure will be 1 – on average, one person was employed. Barring multiple employers, the ILU will measure 12 – a total of 12 persons were employed.

⁵ The Labour Force Survey grand total employment figure for that year was 14.5 mln. The Survey of Employment, Payrolls and Hours (which excludes Agriculture and certain other industries) measured 11.6 mln.

⁶ Payroll employment data by size class are regularly published in the Small Business Quarterly.

⁷ Size classes were: less than 5 ILUs, 5 to 19, 20 to 49, 50 to 99, 100 to 199, 200 to 499, and more than 500.