The Wheel of (Over)Time

Replication code for "The Wheel of (Over)Time" by Jacob Kohlhepp and Robert McDonough.

In the United States public sector, there are many examples where overtime is allocated informally, and overtime earnings are concentrated among a small number of government workers. Is this government inefficiency driven by insider influence, or an efficient reflection of worker preferences? We study the Los Angeles Department of Transportation, where a few traffic officers earned more than $100,000 in overtime over 1.5 years. A constantly rotating list called "the wheel" assigns overtime equally initially, but officers are allowed to informally trade. Using novel daily personnel records, we recover the position of the wheel as well as the time-varying network of potential relationships between officers. Officers are several times more likely to work overtime when they are well-connected to coworkers likely endowed with overtime. Nevertheless, overtime inequality primarily reflects underlying differences in preferences. Informal trading achieves 93.8% of the maximum possible allocative efficiency, or $4.15 million more than random assignment.

Paper: wheel_overtime_draft.pdf

Quick Start

# 1. Open the RStudio project
#    File > Open Project > wheel.Rproj

# 2. Run everything end to end
source("run_all.R")

# 3. Or run the two master pipelines separately
source("run_prep_data.R")
source("run_analysis.R")

Main Entry Points

• run_prep_data.R: builds the intermediate data products and network panels.
• run_analysis.R: runs the estimation, descriptive analysis, event studies, the active 06_01-06_05 simulation block, and the simulation-comparison scripts.
• run_all.R: runs run_prep_data.R and then run_analysis.R.

Logging and Conditional Execution

Logging lives in utils/logging.R.

• Every numbered script writes its own log in logs/.
• The three master runners also write logs: run_prep_data.log, run_analysis.log, and run_all.log.
• Scripts are skipped automatically when their previous log is successful, their expected outputs exist, and none of their dependencies are newer than the logged completion time.
• Set CONFIG$force_rerun <- TRUE to bypass the skip logic.

Pipeline Overview

Data Preparation

Run via source("run_prep_data.R").

Step	Script	Main outputs
01_01	01_01_process_weather.R	data/weather_daily.rds
01_02	01_02_process_holidays.R	data/holidays.rds
01_03	01_03_mk_working.R	data/employee_data.rds, data/pay_data.rds, data/workers_comp.rds
01_04	01_04_mk_expanded_pay.R	data/working_expanded.rds, data/01_04_fornetwork.rds
01_05	01_05_mk_pre_network.R	data/01_05_pre_network_{30,90,180,1000}.csv
01_06	01_06_mk_network.R	data/01_06_panel_working.rds, data/01_06_panel_working_30.rds, data/01_06_panel_working_180.rds
01_07	01_07_mk_map.R	out/figures/01_07_la_street_map.png

Analysis

Run via source("run_analysis.R").

Tier	Scripts	Depends on
2	02_01_mk_estimation_sample.R	01_06 outputs
3	03_01_facts.R, 03_02_lag_check.R, 03_03-03_08 event studies	Tier 2
3b	Existing modern DiD scripts (*_did2s.R, *_sunab.R, *_cs.R)	Tier 2
4	04_01_estimate.R, 04_02_estimate_many.R	Tier 2
5	05_01_display.R, 05_02_validate_valuations.R, 05_03_cartel_demographics.R, 05_04_decomp_pref_network.R, 05_05_labor_supply.R	Tier 4
6	06_01-06_05 simulation scripts	Tier 4
7	07_01_heatmap.R, 07_02_compare_sims.R	Tier 6

Legacy note: 06_99_sim_frontier.R is kept for optional manual runs, but it is no longer part of run_analysis.R. 07_02_compare_sims.R will use data/06_99_sim_frontier.rds only when that file already exists.

Configuration

Project-wide settings live in config.R.

Important fields include:

• log_dir, data_dir, output_dir
• network_windows and network_window_default
• estimation_start and estimation_end
• force_rerun and verbose_logging

Machine-specific locations can be overridden with environment variables such as:

WHEEL_DATA_DIR=data
WHEEL_OUT_DIR=out
WHEEL_LOG_DIR=logs

Project Layout

• config.R: central configuration and helper utilities.
• utils/logging.R: logging and skip logic.
• run_prep_data.R, run_analysis.R, run_all.R: pipeline orchestrators.
• data/: intermediate data files.
• out/figures/, out/tables/: generated outputs.
• logs/: step-level and master-runner logs.
• date-stamped raw data directories such as 20170803_payworkers_comp/ and 20190811_weather/.

Requirements

• R >= 4.3.0
• Core packages include data.table, fixest, circular, ClusTorus, sf, osmdata, ggplot2, lubridate, stringr, almanac, tidygeocoder, and haven