Real Effort Tasks

Laboratory-in-the-field tasks that require real cognitive or physical effort — counting, sorting, encoding — to measure preferences over effort, time allocation, and the disutility of work without relying on hypothetical responses.

What it is

Real effort tasks replace hypothetical work scenarios with actual work: respondents perform a repetitive cognitive or physical task — counting objects, encoding numbers into symbols, pressing buttons in sequence — and are paid based on their output. Because the effort is real and the opportunity cost is real (time spent on the task is time not spent elsewhere), choices about how much to work and when to work reflect genuine preferences over effort, not stated preferences.

Real effort tasks are used in experimental economics to measure the disutility of effort, effort provision under different incentive schemes, time preferences for effort (as opposed to money), and present bias in the domain of work. The core advantage over monetary time preference tasks is that effort tasks tap preferences in the domain most relevant to labour supply and commitment decisions, which may differ substantially from monetary preferences.

When to use it

Real effort tasks are appropriate when the research question concerns effort allocation, labour supply incentives, or the behavioural responses of workers to different compensation structures. They are also used as a method for identifying present bias in the effort domain, which has direct relevance to questions about commitment devices, training completion, and technology adoption.

Augenblick, Niederle and Sprenger (2015) run a real effort task — counting zeros in a matrix — and find strong evidence of present bias in the effort domain that is not predicted by monetary time preference measures from the same participants. This paper is foundational for the claim that effort preferences and monetary preferences are distinct and should be measured separately when effort is the relevant outcome variable. Deserranno (2019) uses output on a real effort task as a measure of productivity in a study of NGO worker selection in Uganda.

The method is less appropriate when the task is so simple that it produces ceiling effects (everyone works at maximum output), or when it is so unfamiliar that performance reflects skill rather than effort preferences.

How it works

A real effort task has three elements:

The task itself. A repetitive activity with a well-defined unit of output that does not require specialised skill — counting, sorting, encoding — so that performance variation reflects effort rather than ability. Common tasks include:

Slider task: Move sliders on a screen to the midpoint of a scale. Simple, fast, and easily computerised.
Matrix task: Identify the two numbers in a 4×4 matrix that sum to exactly 10. Requires sustained attention.
Letter encoding: Transcribe a sequence of numbers into a code using a provided lookup table. Consistent difficulty across rounds.
Physical counting: Count objects (seeds, tokens) placed in a container. Usable without computers.

The incentive structure. Participants are paid per unit of output, per task completed, or in a piece-rate scheme. Varying the piece rate allows estimation of the elasticity of effort to wage. Fixed-time designs (work as much as you want in N minutes) or fixed-output designs (complete K units as fast as you want) are both used.

The intertemporal design. For measuring present bias and time preferences over effort, participants are asked to commit in advance to how many units they will complete in a future session, and separately to decide in the moment how many to complete in the current session. Discrepancies between advance commitment and in-the-moment choice identify present bias — respondents work less in the current session than they committed to, showing that effort aversion is larger when effort is immediate than when it is future.

Key decisions

Task selection. The task should be tedious but not cognitively demanding at the level that performance would vary with cognitive ability. It should also be culturally neutral — number-based tasks assume numeracy; physical tasks may be more accessible but less standardisable. Pilot testing the task in the study population is essential to establish that all participants can perform it at a reasonable rate and that variance in performance reflects effort, not comprehension.

Session structure. Single-session designs measure effort provision at a point in time. Multi-session designs — where participants commit today to work done in a future session — are necessary for identifying present bias. Multi-session designs require tracking participants across sessions and maintaining credibility that the future sessions will actually occur.

Piece rate vs. flat rate. Piece rates create direct incentive to work; flat rates create no marginal incentive for additional output. The comparison between piece-rate and flat-rate conditions — or between different piece rates — estimates the wage elasticity of effort. For measuring time preferences, piece rates are necessary to make the future session valuable.

Payment in real time vs. delayed. In multi-session designs, whether payment for the current session is given immediately or delayed (matched to the future session’s payment) affects the comparison between current and future effort. The design in Augenblick et al. (2015) carefully separates the timing of work from the timing of payment to isolate effort preferences.

Fatigue and learning. Real effort tasks in long sessions produce both learning effects (performance improves as the task becomes familiar) and fatigue effects (performance declines over time). These can be disentangled in within-session panel designs by analysing output by block. For cross-person comparisons, standardising session length and task complexity reduces noise from these sources.

Caveats & common mistakes

Skill confounds. If the task requires skills that vary across participants — numeracy for number tasks, dexterity for physical tasks — then performance differences reflect ability differences rather than effort differences. The task should have a high ceiling so that ability is rarely binding: even a low-ability participant should be able to complete the maximum output if they try hard. Tasks with ceiling effects in performance rates signal that effort is not the binding constraint.

Demand effects in multi-session designs. Participants who commit to future effort in a session with a researcher present may feel socially obligated to follow through, inflating apparent commitment device take-up or apparent present-bias correction. Tracking actual completion in future sessions — rather than relying on stated commitment — is necessary for valid inference.

Domain specificity of results. As Augenblick et al. (2015) establish, present bias in the effort domain does not predict present bias in the monetary domain and vice versa. Results from effort tasks should not be generalised to monetary time preferences, and vice versa. This is a feature — the tasks measure genuinely different things — but it means that effort tasks should be used specifically when the relevant behavioural domain is labour supply or work completion.

Incentive compatibility across tasks. When real effort tasks are run alongside other incentivised tasks (trust game, risk elicitation), earnings from earlier tasks create wealth effects that may affect effort provision. Randomising task order and using a single random payment protocol across tasks reduces this.

Analysis Guide

# units_completed: number of task units completed in the session
# minutes_worked: total session time
# piece_rate: payment per unit (varies across respondents or rounds)
import pandas as pd
import numpy as np
import statsmodels.formula.api as smf
import matplotlib.pyplot as plt

# 1. Compute output per minute — normalising by time is necessary because session
#    lengths vary; this is the primary effort measure and allows comparison across
#    respondents who spent different amounts of time on the task
df["output_rate"] = df["units_completed"] / df["minutes_worked"]

# 2. Plot the effort supply curve by piece rate — if the wage elasticity of effort is
#    positive, output_rate should increase with piece_rate; a flat or declining curve
#    suggests a ceiling effect (everyone already working at maximum) or a piece rate
#    range too narrow to create meaningful incentives
curve = df.groupby("piece_rate")["output_rate"].mean().reset_index()
plt.plot(curve["piece_rate"], curve["output_rate"], marker="o")
plt.xlabel("Piece rate"); plt.ylabel("Units per minute"); plt.title("Effort supply")

# 3. Identify present bias by comparing what respondents committed to in the prior
#    session versus what they actually completed — respondents who complete fewer
#    units than they committed to reveal that effort aversion is larger when effort
#    is immediate than when it is future; above 30-40% is consistent with the
#    Augenblick et al. (2015) evidence on present bias in the effort domain
df["present_biased"] = (df["actual_future_units"] < df["committed_future_units"]).astype(int)
print(df["present_biased"].mean())

# 4. Regress units completed on piece rate and characteristics — the coefficient
#    on piece_rate is the wage elasticity of effort; near-zero means ceiling effect
#    or insufficient incentive range; include demographics as controls
fit = smf.ols("units_completed ~ piece_rate + age + C(female) + log_hh_expenditure",
            data=df).fit(cov_type="HC1")
print(fit.summary())

# 5. Estimate the treatment effect on effort separately — a positive treatment
#    coefficient means the treatment arm completed more units; check whether this
#    operates through output_rate (intensity) or total duration
fit_t = smf.ols("units_completed ~ C(treatment) + age + C(female)",
              data=df).fit(cov_type="HC1")
print(fit_t.summary())

# 1. Compute output per minute — normalises for session length differences and is
#    the primary effort measure; absolute values are not comparable across tasks,
#    so use this as a relative measure within your study
df$output_rate <- df$units_completed / df$minutes_worked

# 2. Plot effort supply by piece rate — a positive slope confirms higher pay increases
#    effort; a flat slope signals a ceiling effect or insufficient incentive range
library(dplyr)
effort_curve <- df |>
group_by(piece_rate) |>
summarise(mean_output = mean(output_rate, na.rm = TRUE))
plot(effort_curve$piece_rate, effort_curve$mean_output, type = "b",
   xlab = "Piece rate", ylab = "Units per minute", main = "Effort supply")

# 3. Identify present bias — respondents who completed fewer units than they committed
#    to in the prior session reveal that effort costs feel larger when they are immediate;
#    prevalence above 30–40% is consistent with present bias in the effort domain
df$present_biased <- as.integer(df$actual_future < df$committed_future)
mean(df$present_biased, na.rm = TRUE)

# 4. Regress units completed on piece rate and characteristics — the piece_rate
#    coefficient is the wage elasticity of effort; positive and significant means
#    effort is responsive to incentives in this population
library(estimatr)
fit <- lm_robust(units_completed ~ piece_rate + age + female + log_hh_expenditure,
               data = df, se_type = "HC2")
summary(fit)

XLSForm / SurveyCTO

Computerised real effort tasks are best implemented in tablet applications rather than standard SurveyCTO forms, since the task requires timed interaction, real-time scoring, and visual presentation of task items. If using SurveyCTO, pre-enter task items as a randomised list and use timer constraints to limit session duration. Record both the number of completed items and the total time taken (using SurveyCTO’s built-in timing fields). For multi-session designs, record the participant’s commitment at session 1 in a text field, load it back via pulldata() at session 2, and record actual completion for comparison.

Reading the output

output_rate (units per minute) is the primary productivity measure. Because tasks differ in difficulty and design, absolute values are not comparable across studies — report the distribution within your sample and use it as a relative measure or as a control variable.
A positive coefficient on piece_rate in the regression confirms that higher pay increases output — the wage elasticity of effort. A coefficient near zero suggests the task has a ceiling effect (everyone is already working at maximum) or that the piece rate range is too narrow to create meaningful incentives.
present_biased == 1 means the respondent completed fewer units in the actual future session than they committed to in the prior session. A prevalence above 30–40% is consistent with present bias in the effort domain, as documented by Augenblick et al. (2015). Very high rates (above 60%) may also reflect follow-up attrition or credibility failures in the commitment mechanism.
For treatment effect regressions, a positive coefficient on treatment means the treatment arm completed more units. Interpret this alongside output_rate to check whether the effect operates through intensity (faster work) or duration (longer engagement).

References

Augenblick, N., Niederle, M., & Sprenger, C. (2015). Working over time: Dynamic inconsistency in real effort tasks. Quarterly Journal of Economics, 130(3), 1067–1115. https://doi.org/10.1093/qje/qjv020

Deserranno, E. (2019). Financial incentives as signals: Experimental evidence from the recruitment of village promoters in Uganda. American Economic Journal: Applied Economics, 11(1), 277–317. https://doi.org/10.1257/app.20170670

Gill, D., & Prowse, V. (2012). A structural analysis of disappointment aversion in a real effort competition. American Economic Review, 102(1), 469–503. https://doi.org/10.1257/aer.102.1.469