Productivity growth patterns in US dairy products manufacturing plants

Abstract

We analyse the productivity growth patterns in the US dairy products industry using the Census Bureau's plant-level data set. We decompose Total Factor Productivity (TFP) growth into the scale and technical change components and analyse variability of plants’ productivity by constructing transition matrices. We observe a cross-sectional dispersion in plant-level productivity growth in the industry. Even though the industry aggregate shows a small TFP growth rate −0.3%, quartile rank analysis shows that while the lowest productivity quartile plants average 1.9% loss in productivity, the highest productivity quartile plants average 1.1% growth annually. Our results show considerable movements of plants in their productivity rank categories overall and across age groups, and we find that the scale effect contribution to TFP growth accounts for about 90% of TFP growth on average in the industry. These plants extract scale efficiencies over technological progress to fuel TFP growth. The youngest plants start with the lowest productivity growth at the initial time period, but they catch up older plants productivity, which present the highest average growth rate through years. This may indicate a ‘learning-by-doing’ process for the industry.

Acknowledgements

This research was supported in part by USDA/NRI Grant No. 03-35400-12949. Research results and conclusions expressed herein are solely the authors’ and do not necessarily reflect the views of the Census Bureau. This article has been screened to ensure that no confidential data has been revealed. We are grateful to the editor and an anonymous referee for their helpful comments and suggestions.

Notes

¹ Some of the relevant empirical studies that explored the link between micro-level and aggregate productivity growth are as follows: for US, Haltiwanger (1997) and Foster et al. (2001); for other countries, Aw et al. (2001) and Yasar et al. (2004).

² The weights are related with the plants’ importance in the industry (Foster et al., 2001).

³ The CM is conducted every 5 years (1963, 1967, 1972, 1977, 1982, 1987 and 1992), samples every manufacturing plant. The ASM continuously samples plants with more than 250 employees. Continuous data exists for large plants and for small plants that are selected to be part of the ASM panel. Small plants have missing information for all years except CM and ASM panel years if the plant is selected to be part of an ASM; therefore, comprehensive time series information on small plants is not available. The complete description of the LRD can be found in McGuckin and Pascoe (1988).

⁴ Four-digit level industry groups for the dairy industry are as follows: SIC = 2021 Creamery Butter; SIC = 2022 Cheese, Natural and Processed; SIC = 2023 Dry, Condensed and Evaporated Milk Products; SIC = 2024 Ice Cream and Frozen Desserts; SIC = 2026 Fluid Milk.

⁵ The variable construction is described in more detail in the Appendix.

⁶ For ease of explanation, we organize our result into five different time periods: 1973–1975; 1976–1980; 1981–1985; 1986–1990; 1991–1995.

⁷ We also present an average returns-to-scale for the industry during the time periods. On average, the dairy products industry follows decreasing returns-to-scale in each rank, ranging from 0.92 to 0.93. We calculate average returns-to-scale by finding the point estimates of the returns-to-scale for each plant and group them according to their TFP quartiles. Then, we take the average for each quartile group.

⁸ This finding is more pronounced for all food manufacturing plants at the two-digit level. For all food plants, while overall productivity growth averages 0.9% per year, the lowest quartile averages −7.2% and the highest quartile averages 8.6% per year (Celikkol, 2003).

⁹ The scale effect contributions to TFP growth are in the negative direction for the plants in ranks I and II and positive for the plants in rank IV.

¹⁰ We also checked the outliers in the productivity dispersion. If the difference in each year is greater than 3× (mean of the differences) then that point is considered an outlier. Based on this consideration, the industry does not have any outliers (Celikkol, 2003).

¹¹ The plants’ age and size variables are described in the Appendix.

¹² The period averages show that the oldest plants have the highest TFP growth during the time periods 1973 to 1975 and 1976 to 1980, and the middle-aged plants have the highest TFP growth in the rest of the periods, except 1986 to 1990 in which the youngest plants have the highest TFP growth (albeit negative).

¹³ Plants in age categories 1 and 2 follow a close and similar pattern with a small gap between each other, specifically with no gap after the mid-1981 to 1985 period.

¹⁴ To set up plant transition matrices, we assign each plant to a quartile group in each year based on its TFP measure. Then, we tabulate plant transition occurrences from quartile q(t) in a year t to quartile q(t + 5) in year t + 5 (Bartelsman and Dhrymes, 1998, for a similar approach).

¹⁵ Other than 75% of plants in age 1 and rank III from 1976 to 1981 and all of the plants in age 1 and rank I from 1981 to 1986.

¹⁶ In particular, all of the plants in the time period 1976 to 1981, 80% of the plants in the next time period, and all of the plants in the 1986–1991 period move downward in their productivity rankings. Similarly, all of the plants in rank III during the last period moved downward.

¹⁷ Baily et al. (1992) find that there are signs of greater mobility among plants in their productivity distribution especially 1980s, the degree of persistence has declined overtime.

¹⁸ Kruger (2004) reports that despite the fact that he found strong persistence in productivity transitions, there are some notable exceptions of industries such as computer and office equipment (SIC 357) and leather gloves and mittens (SIC 315) which experience substantial gains and losses of productivity.

¹⁹ The obvious drawback of the value-added approach is the assumption that material inputs are separable from other inputs and cannot be a source of productivity growth. Most of the food industries are characterized as being materials intensive. The ratio of material cost to value of shipments in the US food manufacturing sector exceeds 60%. In some food-manufacturing industries such as meat products and the fats and oils industries, this ratio is even reaching 80%. Therefore, considering materials as a component of the production function specification impacts the productivity measurement results.

²⁰ This method is used by researchers who use the Census Bureau's LRD to generate the real capital stock series such as Cooper et al. (1999) and Dwyer (1996). See also Becker et al. (2005) for an excellent discussion about plant-level capital construction.