7 Appendix

Data: Definitions and Comparability

The data for the individual industry comparisons were obtained by developing a matched set of industries using the Canadian and U.S. Census of Manufactures for 1929. In 1929, the Canadian census listed 168 separate industries; the U.S. total was 328. In the end, after aggregating industries in both countries, we ended up with 137 industries (127 excluding all miscellaneous categories). In matching industries, we made use of product information for individual industries.

In the main, variables were defined in a similar way in the two countries. The differences, when they exist are not critical to our results. Canada defined value of production as value of products produced, while the U.S. used value shipped. However, the U.S. Census investigated whether its use of value shipped rather than produced made much of a difference, and concluded it did not (Fabricant, 1940).

Materials costs were defined in a similar fashion except for two minor differences. Canada excluded returnable containers and included some contract work; the United States did the reverse. Contract work is relatively unimportant except in the clothing industry. This difference biases our estimates of relative Canada/U.S. total factor productivity downwards.

In the absence of dollar value of capital stock, we resorted to horsepower as a proxy. We chose to compare primary horsepower—the sum of horsepower produced by steam engines, internal combustion engines, water wheels and turbines and electric motors using purchased power. Both countries measured horsepower similarly—as rated capacity rather than power used; however, American coverage is less than 100%. We have made no correction for this since it is likely (though not specifically admitted) that Canadian coverage was imperfect. If this is not so, the lower U.S. coverage once more biases our estimates downward.

Some effort was devoted to checking the accuracy of the capital proxy. We examined the ratio of relative Canada/U.S. capital stock derived from census book values and relative horsepower for the years prior to 1929 when both countries reported book value of capital and horsepower and found horsepower closely proxied capital. We traced the growth of horsepower and U.S. real capital stock (Creamer, Dobrovolsky and Borenstein, 1960) for each decade from 1899 to 1929 and found they moved closely. But perhaps our best evidence is that our estimate of the median Canada/U.S. ratio of output per horsepower is 1.05 (with a mean of 1.16). Latourette (1968) obtains very similar ratios when he builds a capital stock from investment flows.

Similar definitions were employed in both countries for the labour force—though measurement practice differed. In the U.S., the number of production workers was derived by dividing by 12 the sum of monthly figures—the latter being the simple sum of establishment totals for the month. In 1929, Canada did essentially the same but adjusted its measure by dividing the sum by the number of months establishments were in operation—thereby taking into account seasonal operation. This will once more bias our calculated TFP ratios against Canada.

Wage and salary costs were defined similarly in the two countries except that the United States did not include an imputation for income derived by proprietors while Canada did so. Kuznets (1941) estimate for the size of this component suggests this creates little bias. When it comes to establishment definitions, there were some important differences. In Canada, an establishment if defined as an operating unit—basically a single plant; the United States however occasionally grouped closely-located plants. The coverage of establishments in the two countries also differed. In 1929, the United States defined an establishment as one having a value of shipments of $5,000 or more. In Canada, all establishments regardless of size were included. As a result of both differences, our estimates of relative plant size presented in Table 4 are biased downwards.

In order to correct for output price differences between the two countries in 1929, Canadian and U.S. wholesale price data were used to calculate price relatives for that year using a Tornqvist divisia index. Data on U.S. prices came from (1) the Products of Manufacturing Industries, 1929 bulletin published as part of the 15^th Census of the United States, 1930, and in some cases form (2) the Biennial Census of Manufacturing, 1931, which essentially reproduces the information in the Bulletin. Data on Canadian prices were compiled from several different sources. The first choice was always the individual industry studies published by the Dominion Bureau of Statistics. In some cases, we were able to use unpublished information that was not confidential to fill in gaps. As not all industries were covered by these publications, and not all industries had information that was sufficiently extensive for our purposes, it was necessary to make use of supplementary sources. These include 1) the D.B.S. publication Prices and Price Indices and 2) the Canada Year Book, 1932.

Relative Canadian and U.S. wholesale price indexes were drawn up for the 137 matched industry categories. In many cases, we were able to generate relative prices for the components of the subsectors. Two relative prices were calculated using Tornqvist Divisia indexes. The first (RPC) used Canadian quantity weights; the second (RPU) used U.S. quantities as weights. When commodity data were not available for a category, a third price (RPA) was calculated as the average of like industries—generally those within the same 2-digit category. Of the 137 sectors, detailed price information was available for 75 of our matched industries to calculate RPC or RPU. We were only able to match 51 industries for input prices.