The Incidence of and Pay-off to Employer Training: A Review of the Literature with Recommendations for Policy

Chapter Three
Bishop, J.
Cornell University, 1994

3. The Effects of Employer Training on Worker Productivity

3.1 Studies by Industrial Psychologists

Industrial psychologists have conducted many studies of the effects of formal training on job knowledge and job performance. A recent meta-analysis of the literature on management training by Burke and Day¹ found 22 studies of the effects of training on objective learning criteria (generally scores on paper and pencil tests), 39 studies of the effects of training on subjective behaviour criteria (generally performance appraisals) and 11 studies of effects on objective results criteria. Burke and Day calculated a standardized effect size for each training program and for each criterion by dividing the difference between trainees and an untrained comparison group by the within-group standard deviation of the criterion. This result was then adjusted for criterion unreliability by dividing by the square root of criterion reliability. Mean effect sizes varied with the criterion and the type and method of training, but in each of the categories examined mean effect sizes were positive and significantly greater than zero. A number of studies compared the efficacy of lecture presentations, lecture plus discussion and lecture/discussion plus role playing or practice. The mean 'true' effect sizes were respectively 37, 23 and 93 for objective learning criteria and 46, 11 and 34 for subjective behavior criteria.' These findings provide some support for the conventional wisdom that learning is enhanced when trainees get to practice the skills they are being taught.

On the other hand, a number of studies obtained point estimates of training effects that were negative. Sample sizes are typically small, so sampling error could be the cause of these results. When, however, artifacts like sampling error and criterion unreliability were corrected for by Burke and Day², 90 percent "credibility values (the effect size which 90 percent of true effect sizes should lie above) sometimes fell below zero. This finding suggests that, while most formal training programs achieve their objectives of significantly improving job knowledge and job performance, a significant minority do not.

While cumulative reviews of the training literature by Burke and Day and others provide suggestive evidence about which training methods are more effective, the generalizations that can confidently be drawn from this literature are few. Any one issue is addressed by only a few studies, sample sizes are small (most studies compare treatment groups that contain fewer than 40 people), criterions are often of doubtful relevance to establishment profitability and designs are often flawed (random assignment is often absent). The great variability across studies in the estimates of the magnitude of training effects is further evidence of our ignorance. When one considers that probably more than S20 billion dollars is spent annually on the formal training of managers and supervisors, it is quite remarkable that Burke and Day were able to find only 70 studies (with a cumulative sample size of only 7,178 individuals) that met their acceptance criteria. Clearly, a great deal more systematic field research is required.

3.2 Productivity Effects of Prior Training

Most training is informal not formal. What are the productivity effects of informal training? One way to address this question is to hold the job constant and then compare the productivity of incumbents who have different amounts of tenure and prior relevant work experience. Analyses of the U.S. Employment Service's GATB Revalidation data on 31,399 workers in 159 different occupations at 3052 different firms indicates that both have substantial effects on job performance. This analysis is summarized in Table 8.

Findings from Analysis of USES Data: Relative to someone with no relevant work experience, a worker with 10 year of relevant work experience is predicted to be 28 percent more productive during the first couple of years in technical, craft and service jobs and 12 to 15 percent more productive in clerical and operative jobs. The effects diminish as experience increases, but they do not reach zero until 37 years for operatives, 55+ years for craft workers and high skill clerical workers and 19-31 years for other occupations.

Productivity rises even more rapidly as tenure at the job increases: After ten years on the job, productivity had risen 84 percent in technical jobs, 68 percent in high skill clerical jobs, 62 percent in craft jobs, 45-47 percent in operative and service jobs and 32 percent in low skill service jobs? The effect of tenure on job performance stops rising and starts to decline at somewhere between 16 and 24 years of tenure.

Except for technicians, age (interpreted as general experience) has large effects on job performance as well. Holding tenure and occupational experience constant, being ten years older (28 rather than 18) raised productivity 8 percent in low skill clerical jobs, 17-18 percent in operative and service jobs, 23 percent in high skill clerical jobs and 33 percent in craft jobs.

Findings from Analysis of NFIB data: A survey of a stratified random sample of the 700,000 member National Federation of Independent Business (NFIB) during the first half of 1987 provides another source of information on the productivity payoffs to training. About 1400 firms provided information on the training received and the productivity of two recently hired workers occupying the same job. By analyzing the determinants of the differences in productivity and wage outcomes for these two workers, one can assess the impacts of training received on previous jobs and in schools on productivity and wage rates (Bishop 1994).

Relevant work experience significantly increased the productivity of new hires and significantly reduced the time required to train them. Holding total experience constant, new hires with ten years of relevant experience required less training, higher productivity (20 percent at the end of six months and 13 percent at the time of the interview), made more suggestions for improving productivity and were paid 22 to 25 percent more.

Total work experience was defined as the total number of years since completing school or reaching age 16 whichever is smaller. In the NFIB data, experience that was not relevant to the job did not have positive effects on productivity and retention. Ten years of irrelevant experience, in fact, reduced productivity at 6 months of tenure by a statistically significant 7 percent. Even though it is associated with lower productivity, irrelevant experience is associated with higher wage rates relative to co-workers. The effect of irrelevant experience on the wage is about one-third of the size of the effect of relevant experience on wage rates.

Table 8: Within-Job Productivity Effects of Previous Work

Note: Table entries are estimates of effects of a particular form, of experience while holding other forms of experience, schooling, test scores, gender and ethnicity constant. The productivity effects are expressed as a percentage of the mean level of compensation in that occupation. Estimates of productivity effects in 1985 dollars may be obtained by multiplying the percentage reported in rows 1 to 3 by the mean compensation in row 4. These estimates hold the employer and the job constant and thus capture only a portion of the benefits of on-the-job learning. Learning also helps the individual enter higher paying occupations and obtain jobs at better paying companies and these effects are best measured by regressions predicting wages in representative samples of the population.
Sources for Table 8: Derived from an analysis of the U.S. Employment Services Individual Data File developed for revalidating the General Aptitude Test Battery. Deviations from the mean performance rating for the job were regressed on deviations from the mean values (for the job) of schooling, test scores, gender, race, Hispanic and tenure, total occupational experience, and age and their squares. Details are available in Appendix C of Bishop³. The higher rates of turnover of unsuccessful employees cause selectivity bias watch was corrected for by multiplying the coefficient in table C1 of Bishop 1994 by 1.76⁴ (Goldberger 1981). The result was then multiplied by estimates of the standard deviation of output across workers taken from Bishop⁵. True effects are probably larger because both independent and dependent variables are measured with error and this causes our estimates to be biased toward zero. No correction was made for measurement error bias.

Employers were pleasantly surprised by the productivity of workers with relevant work experience and unpleasantly surprised by the productivity of those with irrelevant work experience. These findings suggest that many employers were not aware of the relevance of the new hire's previous work experience until long after the hiring decision.

Formal training received on-the-job from a previous employer increased initial productivity by 9.5 percent of the wage and reduced training requirements by 17 percent. It had no effect, however, at the time of the interview. Formal training received off-the-job, on the other hand, had no effects during the first few months at the firm, but it increased the index of suggestions by 37 percent and productivity at the time of tae interview by 15.9 percent. Formal off-the-job training does not increase current wage rates, !however, so profitability increased by 13.8 percent of the wage at six months of tenure and by 18.6 percent of the wage at the time of the interview.

These results suggest that on-the-job training sponsored by firm A not only benefits the employee and employer (as implied by Becker's theory of OJT), but also sometimes benefits other employers in the industry who hire workers who quit or are laid off by firm A. Formal off-the-job training generates substantial long lasting externalities (benefits received by the worker's future employer: and by consumers). The informal training proxied by the relevant experience variable appears to generate externalities only in the first year or so of a worker's tenure at a firm.

3.3 The Impact of Training on the Productivity of New Hires at the Training Firm

The big improvements in productivity during the first year on a job suggest that the total rates of -return (combining both worker and employer benefits and costs) are likely to be very high during the first months of employment. For clerical workers, for example, the total costs of training during the first 3 months were 235 hours or .113 of a year's output by a worker whose skill level was equal to that of a new employee. Since this figure is an upper bound on the investment that contributed to the 40 percent gain during the first months on the job, the average rate of return must have been above 354% per year (.40/.113). Since the intensity of training investment falls with tenure at the firm, the cost of training investment during the next 21 months cannot have exceeded .7875 (i.e.1.75*235/520) of a year's productivity by a newly hired worker. This implies that the average rate of return to training investments during this 21 month period exceeds 40% per year (32/.7875).

However, marginal rates of return to training investment are lower than average rates of return and some of the gain in productivity results from learning by doing and not from training. Multivariate cross section models of productivity growth are necessary to tackle the issue of the marginal productivity of on-the-job training.

Multi-variate analyses of the effects of training on rates of improvement in productivity of new hires⁶ have found that:

• Hours devoted to each type of training had very similar effects on productivity growth during the first year or so on the job. This implies that lower cost forms of training - informal training by coworkers and training by watching others do the job - had higher benefit cost ratios than informal training by management and formal training.
• Formal training had significantly larger effects on wage growth than informal training. Formal training's tendency to have larger effects on wage growth and quit rates than informal training probably result from the fact that formal training is better signalled to the labor market.
• The productivity growth effects of formal training were bigger at large establishments.
• The reported generality of on-the-job training had no significant effects on its marginal productivity.
• When training was reported to be highly general, training had a larger effect on wage growth than when training was reported to be specific. Nevertheless, training that was reported to be entirely general had much larger effects on productivity growth than wage growth implying that the labor market treats this training as if it were at least partly specific to the firm.

The multivariate analysis of EOPP data presented in Bishop⁷ generated tentative estimates of both the opportunity costs and the productivity effects of training (general and specific worker and firm financed combined). These estimates allow a calculation of the marginal gross rates of return (for general and specific training combined) necessary to cover the cost of capital, losses due to turnover and obsolescence. The data were not collected for this purpose, however, so there were gaps that could only be filled by some judicious assumptions.' Consequently, the estimates of marginal gross rates of return (Marginal GROR) for each form of training that are reported in Table 9 must be viewed as very tentative results which will hopefully be displaced when better data sets become available. Marginal GRORs are the ratio of the increment to yearly productivity generated by a small increase in training divided by the cost of increased training (A detailed description is in the notes of the table).

The estimated marginal gross rates of return diminish as the intensity of training increases. The mean training intensity for the fast 3 months expressed in units of the time of trained workers is 148 hours. As intensity during the first 3 months rises from 100 hours to 300 hours (double the mean), the marginal gross rate of return for informal OJT by co-workers drops from 43-45 percent to 25-32 percent in the two linear models for typical new hires presented in Table 7 of Bishop 1991. The linear model's GROR drops from 38-43 percent to 25 percent for watching others and from 17-23 percent to -1 to 10 percent for training by supervisors. The marginal GROR of formal OJT is estimated to drop from 11-15 percent at, 100 hours to -3 percent at 300 hours. Estimated gross rates of return calculated from models based on logarithmic specifications are considerably higher than those based on linear specifications of productivity growth. Gross rates of return are also typically higher for the models using the logarithm of training intensity and the square of this logarithm presented in Bishop (1991 Table 8). At the training intensities that typically prevail during the first quarter, marginal gross rates of return are often above 40 percent.

Table 9: Estimates of Marginal Gross Rates of Return Estimates

Estimates of the marginal gross rates of return to increases in the intensity of training at two different levels of training intensity. a 100 hour investment during the first quarter of the job and a 300 hour investment during the fast quarter on the job. Hourly cost factors are assumed to be 1.8 for formal training, 15 for training by supervisors, 1.0 for training by c coworkers, and 0.8 for watching others. When productivity growth over 2 years for the typical individual is being modeled, duration adjusted cost factor is calculated by multiplying by the hourly cost factor by 3 for the reasons given in the text. When productivity growth of a particular individual during the first 14 months is modelled, the duration adjusted cost factor is calculated by multiplying the hourly cost factor by 22. The results presented in the fast panel are calculated by taking the derivative of the estimated regression equations with respect to hours of the specified kind of training, then multiplying by 2000, the assumed number of hours worked in a year, and then dividing by the duration adjusted cost factor. As an example of the calculation, the formula for formal O1'r using the coefficients from the linear model for training intensity (T) equal to 300 hours was as follows:

[(.00046 - .00000049"T-2-L8)'2000 ] / [3'1.8] = -.0256 and the co-worker training formula is:

[(.00077 - .00000049.'T-2)•2000]/[3] - 3173. (Note that the coefficients must be divided by 100 and 10000 in order to scale them in hours of training). The GROR estimates presented in the second panel assume that the firm has 185 employees (this zeros out the 5th and 7th terms of equation 3) and that all of the training received is of the type indicated. For informal training by supervisors, the formula is:

(b2 + b,'1nT`2)'2000/(T•duration factor) which is [(.003 +.0064'4.605`2)'2000] / (100`3) =.4176 at T=100 for the linear productivity growth model for typical workers. For training by watching others, the formula is N + b3f + b,`lnT`2)'2000/(T'duration factor) which is [(.003 + .013`Sw +.0064`4.605'2)`2000] / (100'3) =504.

Obsolescence of skills and turnover mean that these cash flows do not have an infinite duration and should therefore be compared to the sum of the interest rate, the obsolescence rate and the turnover rate times the proportion of skills that are effectively specific to the firm.

It must be remembered, however, that these marginal GRORs include cash flows necessary to compensate for turnover and obsolescence and are, therefore, not directly comparable to the real rates of return to schooling and financial assets that typically lie in the range from 5 to 10 percent. If all training investments are specific to the firm and must, therefore, be written off if workers leave and rates of turnover are high, first year GRORs of 30 percent or more will be required to induce the firm to invest in specific training Lillard and Tan⁸ have estimated that the wage effects of formal training depreciate (either due to obsolescence or changing jobs) at 15 to 20 percent per year. This also would imply that equilibrium in the training market would likely yield marginal GRORs of 30 percent or more. Tan et. al.⁹, however, estimates a much lower depreciation rate for wage rate effects of company training - 6 to 7 percent per year. With all the uncertainties regarding the best specification of the productivity growth model, measurement error in the training variables, the specificity of the training, turnover rates, and the obsolescence rates, robust estimates of net rates of return to on-the-job training comparable to rates of return on financial assets and physical capital are not now feasible and will not become feasible until much better data sets become available.

3.4 Organizational Effects of On-the-Job Training

For most kinds of training, outcomes are as much organizational as individual. After reviewing studies of the effect of OJT on organizational productivity, Kochan and Osterman¹⁰ concluded that:

These studies provide consistent and convincing evidence that

1. education and training are associated with significant productivity increases when their impact is examined in a production function context; and
2. training and associated flexible human resource systems are associated with higher levels of productivity and quality in matched comparisons.

The number of studies is quite limited, however. Summarizing his study of flexible manufacturing in Japan and the United States, Ramchandran Jaikumar's¹¹ concluded that:

The heart of this new manufacturing landscape is the management of manufacturing projects: selecting them, creating teams to work on them, and managing workers' intellectual development.

Marcie Tyre's¹² examination of several plants in a single multi-national corporation found that the American plants took longer to start up and had flatter learning curves than plants in Italy and Germany. She attributed this in part to less development and cross-training of workers. A study of hot-roll steel facilities by Ichniowski, Shaw and Prennushi¹³ found that plants using high performance work systems had less down time and produced higher quality output. Higher levels of training were one of the components of the high performance work systems that generated these positive outcomes. Studies of the auto industry by MacDuffie, Krafcik and their colleagues¹⁴ came to similar conclusions. Case studies of plant level productivity in five industries - clothing, kitchen cabinet making, biscuit manufacturing, tool making and hotels - conducted by researchers associated with the National Institute of Economic and Social Research found that the British companies were less productive than their German and Dutch counterparts and concluded that the quantity and quality of occupational training received by young-workers entering the industry were one of the causes of the differentials¹⁵. In none or these studies, however, were the data sets large enough to allow econometric estimation of the unique causal effects of training holding other elements of the human resource system constant.

The studies cited above establish an association between training, high performance work systems and greater productivity. They are consistent with the proposition that modernization and training are complementary - i.e. training is often critical to the implementation of new technology or a reorganization and, therefore, companies that are modernizing are more likely to be investing heavily in training. This research does not imply that modernization is the only occasion where training is worthwhile. Nevertheless, these studies are sometimes over interpreted as implying that: "Firms that are unwilling to upgrade production technologies and management methods are not ready to train"¹⁶. This statement is not justified by the evidence. Surely, old style construction contractors still need to train the inexperienced carpenters they employ? Surely, firms with a sexual harassment problem need to train supervisors about company policies in this area?

The studies reviewed in section 3.1 through 3.3 have established that traditional employer provided training raises individual productivity and wage rates. Most of the training incidents in these studies were not occasioned by modernization or a TQM reorganization. Taken altogether the economic literature on training suggests that, as long as the company is initiating and paying for training, one can be pretty confident that most of these investments are profitable both for the worker and the firm.

On-the-job training subsidized by the government has a somewhat more spotty record. Programs which arrange and subsidize OJT for disadvantaged workers raise the wages of adults who participate, but appear to lower the wages of youth (under age 22) who participate. On the other hand, analysis of a very small sample of JTPA trainees in the NFIB data suggests that stigma may be biasing these evaluations and that firms that hire JTPA trainees may be getting better employees than expected and paying them lower wages than is typical for the job.

Occupational training of youth in high schools and community colleges and vocational-technical institutes does raise earnings particularly of women. Adults and incumbent workers, however, do not benefit from the occupational training they get at schools when their employer does not contribute to it's costs.

4. Why Do German and Japanese Workers Receive More Training Than American Workers?

American employers appear to devote less time and resources to the training of entry level blue collar, clerical and service employees than employers in Germany and Japan¹⁷. In the automobile industry, for example, newly hired assembly workers receive 310 hours of training in Japan and 280 hours of training in Japanese managed plants located in the US, but only 48 hours of training at US owned plants in the US¹⁸.

Averaged over all auto assembly workers, annual training time is nearly three times greater in plants located in Japan and about 80 percent greater at Japanese plants located in the US. These differentials in training are one of the reasons why Japanese plants are more productive than American plants and Japanese built cars have such a reputation for quality. German employers train their youthful app entices much more thoroughly than American employers train their teenage workers. One visible manifestation of this is the sales personnel one encounters in Germany. They are generally much more knowledgeable about the products they are selling than American sales clerks.

The Japanese and German economies apparently generate a significantly larger number of jobs which offer substantial training on-the-job. Why does this occur? This section of the report reviews the evidence on the culpability of six prime suspects: high turnover, high costs of capital, loose labor markets, lower trainability of American workers, lower rates of technological progress, and the absence of government sponsored signals of skills obtained from training on-the-job.

4.1 Turnover

If American employers were asked why they do not provide more intensive training to young workers, they would probably point to the high turnover rates of youth as the primary reason. And indeed, while some American workers stay at their employer for many years, most workers change employers very frequently. In the 1980s and early 1990s only 383 to 40.5 percent of American workers had been on their current job for more than 5 years. With the exception of Australia and the Netherlands, no other nation had such a low proportion of long tenure employees. The comparable proportions were 63-67 percent for Japan, 59-63 percent for Germany, 58 percent for France, 45 percent for Canada and 45-50 percent for the United Kingdom. For American workers with less than one year of tenure, the probability of a separation in the next 12 months is 59 percent. Since comparably defined turnover is only 32 percent in France, 20 percent in Germany and 24 percent in Japan, national differences in turnover could be a major reason for the low levels of training investment in the US, if the employer's explanation is right¹⁹ .

Turnover effects the stock of trained workers in three ways. First, high turnover necessarily implies that a given rate of investment in firm specific skills yields a smaller stock of workers with firm specific skills. Many of those trained have moved on to other firms where the firm specific components of training yield no benefits.

Second, turnover has a powerful effect on employer decisions to provide training to employees. Employers, not workers, finance most of the training that is undertaken in U.S. firms. Employers will not invest in training unless they believe it will generate a monthly return that exceeds the sum of the monthly turnover rate (generally above 2% per month in the US and sometimes greater than 8%/month) and the cost of capital (which is about 1.5 percent per month or 18% year). Monthly turnover rates are typically much larger than the cost of capital and are also more variable. If turnover is 5% per month and the cost of capital is 1.5% per month, the cash flow yield of the training investment rate of return must exceed 78 percent per year if the investment is to make economic sense. Even when turnover is a very low 2 percent per month, the required cash flow yield is still quite high: 42 percent per year. Training, thus becomes a sensible investment for an American employer only when it yields very rapid and very large returns. The amount of training employers are willing to finance is negatively related to the projected turnover rate of the trainees.

The third reason why turnover is so critical is its impact on the process of teaching and learning. Turnover disrupts learning regardless of whether the skills being learned are generic or firm specific. Schools teach general skills and follow a common curriculum, yet have great difficulty when students transfer from one school to another during the school year. Teaching must be adjusted to the special needs of the learner, and it takes time for the teacher to learn of those special needs. Learning occurs best when instructor and learner have a close personal relationship and it takes time to build such relationships. Turnover is thus one of the determinants of the efficiency of the learning production function.

The high rates of turnover in America, then, help explain why investments in on-the-job training are lower in this country than in Japan and Germany.

[Top]