Soviet Postwar Economic Growth and Capital Labor Substitution

Abstract

What I take exception to in Earl Brubaker's comments is not so much what he says as how he says it. If he were to have phrased things a little more moderately I might almost have agreed with some of the things he said. But the tone of finality about a subject which hardly lends itself to final judgements of that sort may give his remarks an undeserved impression of substance which ought to be removed. Those parts of Brubaker's comments which touch upon difficulties of aggregation, inaccurate measurenment of true inputs, etc., are really universal complaints about using aggregate data that we have heard many times before. I fully and unconditionally agree that they throw up serious obstacles to the measurement, interpretation, and application of aggregate production functions. I thought that this attitude was spelled out clearly enough in my paper. But what else can you do? If y ou want to get an overall picture you've got to use aggregate statistics and aggregative models, flawed as they may be in theory or practice. As Robert Solow has quipped, It may be crooked, but it's the only wheel in town. Brubaker has not inhibited himself from being a heavy consumer of highly aggregated numbers in order to talk about and interpret overall performance. And who can fault him for that? But isn't it then a little unfair for him to make such an issue about this very point in criticizing someone else's work? Then there is the old bugaboo about Soviet Statistics. According to Brubaker can only ponder in awe how these difficulties (of production function estimation) may be compounded where the data have their underlying basis in the statistical and economic systems. This strikes me as an unwarranted exageration. There are some added real problems associated with (today's) statistics, as I indicated in the Data Appendix to my paper. But there may also be areas of equal weight in which their statistics are more accurate and comprehensive than ours (not any worse would really be a more apt description). For example, in compiling capital stock estimates, as I indicated in my paper, the Central Statistical Institute has access to such data as the value of unfinished construction, the value of capital retirements, and value of capital stock as ascertained by periodic inventories. None of these are available for the United States. The lack of unemployment in the USSR, as I also pointed out, eliminates what is one of the major headaches for production function estimation in the West. I certainly don't want to be driven into claiming any more for the data I used than that to a first approximation they are no more unfit for direct production function estimation than U.S. data (this is really damning by faint praise!). Brubaker doesn't like the fact that the imputed factor shares based on the parameter point estimates don't correspond closely enough to Western empirical factor shares. Now it's one order of magnitude to sin against economic theory (as I have) by postulating an aggregate production function and setting about measuring it. It's yet another level of sin to invoke a marginal productivity theory of distribution on one big production function and then identify factor shares with parameter values, especially for the USSR. What's more, there's nothing sacred about the derived parameter point estimates-I certainly wouldn't be willing to go to the stake for them. If my moclel has any relation to reality it has to be a very rough one. While the statistical tests performed seemed to in* Associate professor of economics, Yale University.