Making Do with Less: Working Harder during Recessions

GAICIL Insights

Recessions should make workers less productive.¹ Morale falls, demand shrinks, firms flounder. Yet productivity often rises in downturns, a puzzle that has long nagged economists. Making Do with Less: Working Harder during Recessions, by Edward P. Lazear, Kathryn L. Shaw and Christopher Stanton, separates two explanations. Sorting: the workforce shrinks non-randomly, so the average remaining worker is of higher quality. Effort: each worker produces more while quality is held constant. The authors call the second effect making do with less, getting more effort from fewer workers.

Using individual worker productivity data from a large firm spanning June 2006 to May 2010, the authors measure both effects. Effort dominates sorting for this firm. Productivity rose during the 2007-9 recession because each worker produced more output than in normal times. Labor quality changed little despite substantial turnover. Where the recession hit hardest, productivity gains were strongest and the increase in effort greatest. The relationship holds beyond recession years: cross-sectional increases in unemployment are associated with higher worker productivity. Less productive workers appear most responsive, perhaps because their alternatives shrink the most in weak economic times.

The mechanism is outside options. When alternatives disappear, workers work harder. Output per worker rose because the threat of job loss, or the value of keeping a job, increased. Effort is an economic variable, and it moves with the labor market.

That finding plants a landmine under the productivity statistics everyone will use to judge artificial intelligence. If productivity can jump because workers fear for their jobs, then measured gains during periods of economic stress may reflect effort as much as technology. Imagine AI adoption coinciding with a soft labor market: productivity rises, and the headlines credit the algorithms, when part of the gain is anxious people working harder. Separating AI's contribution from effort and sorting requires individual-level data like this paper's. Without it, policymakers may misattribute productivity changes and draw the wrong lessons.

The paper's method is a ready-made template for evaluating AI inside firms. Panel data on individual output let researchers distinguish productivity changes due to who is employed from changes due to how much each person produces. When firms adopt AI tools, the same decomposition can reveal whether gains come from better tools, a reshaped workforce or shifts in effort. That three-way split is the difference between knowing AI raises productivity per worker and merely guessing.

Distribution enters too. Less productive workers responded most to weak labor markets. If AI alters job prospects, its effect on outside options will vary across workers, and those with fewer alternatives may adjust effort most. The interaction of technology with labor market conditions will shape who bears the burden of transition.

Resilience is the quieter lesson. Firms can maintain or raise output with fewer workers in downturns, partly through effort. AI may strengthen that capacity by helping firms sustain output during disruptions. Policies that support worker transitions, training and job search assistance, help displaced workers find new positions in an economy where firms can produce more with less.

The broader point concerns micro foundations of macro patterns. The rise in aggregate productivity during recessions emerges from individual decisions about effort, shaped by labor market conditions. Understanding those decisions requires detailed data from inside firms. As AI transforms work, the same approach will be needed to untangle how technology, effort and workforce composition combine.

A provocative corollary follows for anyone celebrating early AI productivity numbers: check the unemployment rate first. A gain that coincides with labor market slack may owe something to fear. The paper shows careful measurement can resolve longstanding puzzles. Applied to AI, it will help separate genuine technological gains from cyclical effects, and support policies that foster productivity growth while helping workers adapt.