Valuing Diffuse Global Public Goods from Satellite Constellations: Evidence from GPS and Airline Delays

Lev Ricanati

REVIEW 4 major objections 5 minor 2 references

Valuing Diffuse Global Public Goods from Satellite Constellations: Evidence from GPS and Airline Delays

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Disabling Selective Availability in May 2000 cut airline in-flight delays by 0.617 minutes per flight and produced $268 million in first-year passenger welfare.

desk verdict A transparent but under-identified DiD that quantifies a real gap in the GPS valuation literature; the $268M headline overstates what the paper's own specifications support. read the letter →

arxiv 2506.08209 v1 pith:IMTRVTYK submitted 2025-06-09 econ.GN q-fin.EC

Lev Ricanati This is my paper

classification econ.GNq-fin.EC

keywords GPSSelectiveAvailabilityairlinedelaysdifference-in-differencesglobalpublicgoodsvalueoftraveltimewelfareanalysissatellitenavigation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that a discrete accuracy improvement to a diffuse global public good produced a large, measurable welfare gain. When the United States disabled Selective Availability on May 2, 2000, civilian GPS accuracy jumped tenfold almost overnight, and the paper estimates that US domestic flights experienced 0.617 fewer minutes of in-flight delay per flight on average as a result. Scaling this saving by monthly passenger counts and federal estimates of the value of travel time gives $268,123,934 (2000 dollars) in consumer welfare in the first year, more than double the amount Congress appropriated for GPS that year. The result matters because it shows how small efficiency gains in a non-rival, non-excludable system can aggregate into hundreds of millions of dollars, and it provides a benchmark for the welfare losses now threatened by GPS jamming and spoofing.

What carries the argument

The analysis is carried by a difference-in-differences regression in which the outcome is adddelay (arrival delay minus departure delay), the treated group is all flights in 2000, the control group is all flights in 1999, and a GPS dummy marks flights from May 2 through December 31. The coefficient on the interaction of treated and GPS, reported as -0.617 minutes per flight, is interpreted as the causal effect of the accuracy improvement on in-flight delay. The welfare calculation multiplies the minutes saved by monthly domestic passenger counts and by the federal value-of-travel-time estimate of $28.60 per hour (with lower and upper bounds of $23.80 and $35.60), converting the time saving into dollars.

What would settle it

Run the same difference-in-differences regression with a placebo treatment cutoff, such as May 2, 1999 or any randomly chosen date in the middle of 1999; if the interaction coefficient again lands near -0.6 and is statistically significant, the 0.617-minute estimate is an artifact of seasonal or year-specific variation rather than the Selective Availability change.

Watch

Extended reading notes

Core claim

The central discovery is that removing Selective Availability reduced in-flight delays — measured as arrival delay minus departure delay, or adddelay — by an average of 0.617 minutes per flight in the year following the policy change. The estimate comes from a difference-in-differences regression comparing flights in 2000 (the treated group) with flights in 1999 (the control group), before and after May 2, with airport cancellation rates as weather controls. The effect was not uniform: flights over 2,000 miles saved about 2.4 to 2.9 minutes per flight, and the per-flight saving first grew to more than 1.5 minutes in the early months after the change before declining toward the yearly average. The paper interprets the mechanism as pilots using the more accurate signal to adhere more closely to their flight plans.

Load-bearing premise

The estimate rests on the assumption that, without the GPS change, the seasonal pattern of in-flight delays from January through April to May through December would have been the same in 2000 as it was in 1999; with only two years of data this parallel-trends assumption cannot be tested.

Editorial extensions

If this is right

If the estimate is correct, the $268 million first-year welfare gain from US airline passengers alone exceeds the GPS program's annual appropriation, so the removal of Selective Availability was welfare-enhancing on this margin alone.
Larger savings on flights over 2,000 miles support the proposed mechanism that more accurate GPS lets pilots follow flight paths more closely, since longer routes have more distance to optimize.
The per-flight saving was largest in the first few months after the change and then declined toward 0.617 minutes, suggesting pilots and airlines initially captured larger gains before incorporating the improved signal into standard procedures.
The same method provides a template for pricing GPS disruptions: at 25 minutes of added delay per affected flight, the roughly 41,000 spoofed flights documented in a single month would imply about $110 million in passenger welfare losses.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural falsification exercise would be to run the identical regression with a placebo treatment date, such as May 2, 1999; the paper does not provide one, and such a test would reveal how much of the 0.617-minute coefficient reflects ordinary year-to-year seasonal variation.
The paper's preliminary look at ticket fares suggests airlines may have raised prices after the accuracy improvement; if so, part of the measured welfare gain was captured by producers, and the consumer-surplus figure of $268 million is an upper bound.
The same difference-in-differences design could be applied to other GPS-reliant activities with time-stamped outcomes, such as maritime transit times or logistics fleets, to test whether the delay reduction is specific to aviation or a more general accuracy effect.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The useful thing here is the question, not yet the answer. This is the first paper I know of that tries to put a dollar value on the aviation delay savings from turning off Selective Availability in May 2000. The prior GPS valuation reports (Pham 2011, O'Connor et al. 2019) explicitly left aviation out, so the gap is real. The author uses public BTS data, documents the construction of the adddelay variable, and reports a large set of specifications in the appendices. That transparency is genuine and should be credited. The distance-gradient result — larger effects on longer flights — is also plausibly mechanistic and worth taking seriously as a hypothesis.

The soft spots are in the identification and the framing. The design is a two-year DiD with no control group; 1999 is the only counterfactual for 2000. The parallel-trends assumption is untestable with one pre-period, and the paper's own windowed regressions do not help. The interaction coefficient runs from -0.434 (one month) to -1.687 (three months) to -0.617 (full year). The author reads this as a recency effect, but that is a post-hoc story. The same pattern is exactly what you would see if the year-over-year seasonal baseline shifted unevenly across months. The stress-test suggestion is right: a placebo regression on 1998 vs 1999 with the same May 2 cutoff would adjudicate, and the paper does not provide one.

The headline "at least $268 million" is not supported by the paper's own numbers. The fullyear regression gives $183M, with a lower-bound valuation of $153M. The $268M comes from the most segmented specification, selected after running many variants, and its lower bound is $223M. Calling the largest estimate "at least" inverts the direction of the uncertainty. The welfare calculation itself is straightforward and the DOT value of time is an external input, so no circularity there — the problem is that the causal coefficient feeding into it is fragile.

Who gets value from this paper? Someone working on GPS economic impact who wants a new line of evidence and a clear template for the data construction. As a policy claim, the central number should not be used without substantial caveats. I would send it to a referee, because the question matters and the author has done enough careful work that a serious referee could push it into a much more credible shape. The placebo test needs to be the first requirement.

Referee Report

4 major / 5 minor

Summary. The paper estimates the welfare effect of the United States' May 2, 2000 decision to disable Selective Availability (SA) on GPS accuracy. Using BTS Airline On-Time Performance data for 1999 and 2000, the author constructs a difference-in-differences design in which 1999 flights serve as the control group and 2000 flights as the treated group, with the 'post' period defined as May 2 through December 31. The outcome is adddelay (arrival delay minus departure delay), and the key coefficient is the interaction in Equation 3, reported as -0.617 minutes per flight in Table 4. The author multiplies this time saving by T-100 passenger counts and DOT value-of-time estimates to obtain a headline welfare gain of $268,123,934 in the first post-treatment year. Additional specifications split the sample by recency to treatment and by flight distance, and the paper extrapolates the method to estimate welfare losses from recent GPS jamming and spoofing incidents.

Significance. If the causal estimate were credible, the paper would fill a genuine gap by quantifying a discrete improvement to a global public good using a natural experiment. The manuscript is transparent about data sources, uses public microdata, and applies an externally supplied value of time rather than fitting welfare parameters internally. The historical policy episode is well-motivated and the aviation channel is plausible. However, the central identification assumption is untestable with the two years of data used, and the paper's own estimates across alternative windows are mutually inconsistent by a factor of roughly three. Because the headline welfare number depends on selecting the most disaggregated specification, the quantitative contribution cannot currently be distinguished from specification search and annual seasonal drift.

major comments (4)

[Eq. (3) and Table 5] The identifying assumption for the difference-in-differences coefficient is that the seasonal pattern of adddelay from January-April to May-December was identical in 1999 and 2000, since 1999 serves as the sole control year. This assumption is untestable with only two years of data, and the paper's own estimates contradict it. The interaction coefficient in Table 5 is -0.434 for the 1-month window, -1.508 for the 2-month window, -1.687 for the 3-month window, -1.185 for the 4-month window, and -0.617 for the full year. These specifications all estimate essentially the same first-year treatment effect under a stable seasonal baseline, so a 3.5-fold range indicates that year-over-year baseline shifts are month-specific rather than constant. The author should provide a placebo test using 1998 versus 1999 with the same May 2 cutoff, or another falsification exercise, before the coefficient can be interpreted causally.
[Sec. 4.2 and Appendix C.4] The headline welfare estimate of $268,123,934 is not the estimate from the pre-specified full-year regression in Table 4 ($183,464,122) but the largest estimate among four sets of welfare computations, produced by the most granular specification in Appendix C.4 in which coefficients vary by month and distance group. The paper reports $183M, $248M, $204M, and $268M for four scenarios and selects the maximum as the central result, without a principled model-selection rule or pre-registration. This selection materially inflates the headline number; the abstract's 'at least $268 million' is therefore not supported by the full set of estimates.
[Sec. 4.1, Table 6, and Fig. 4] The mechanism story is not independently tested. The paper interprets the large coefficients for flights over 2000 miles (e.g., -2.883 minutes in Table 6) and the non-monotonic recency pattern as evidence that pilots used improved GPS to adhere to flight plans, with a post-hoc 'learning then regulation' narrative in Section 4.1. No data on GPS receiver equipage, pilot training, or route changes are presented, and the same pattern could arise from distance-specific seasonal shocks or from the extreme sensitivity of the estimates to the chosen window. The mechanism should either be tested with auxiliary data or presented explicitly as speculation.
[Sec. 3.1] The paper acknowledges the absence of a natural control group ('Due to the lack of a second United States to use as a control'), yet the entire research design relies on the assumption that 1999 is a valid counterfactual for 2000. The two-year structure means any year-specific shock that differentially affects the May-December period (weather, fuel prices, air traffic volume, or airline scheduling practices) is absorbed into the interaction term. The author's assertion that 'it appears that there was not a significant alteration in how air travel operated' is not supported by any evidence; this is a load-bearing point that needs direct testing or a bounding exercise.

minor comments (5)

[Fig. 1] The claim that the graph 'appears to show more pronounced valleys and peaks before GPS SA was disabled' is not supported by simple visual inspection; a formal seasonality comparison would be needed to substantiate this.
[Table 3] Table 3 does not list the control-period dates for the recency-based regressions; the text should specify that the control periods are the corresponding calendar windows in 1999.
[Table 14] In the May '00 row, the 2000-2500 miles entry reads '8 1.361', which appears to be a typographical error for '-1.361' based on the corresponding estimate in Table 15; please correct.
[Sec. 4.1] The text reports p values 'less than 0.00'; this should be written as p < 0.001 or with exact values, since p values cannot be zero.
[Appendix C.4] The welfare calculation for the 'most specific' specification replaces one insignificant coefficient with a distance-based coefficient, but the choice of replacement specification is not justified; this should be documented and justified.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the welfare estimate is a product of an externally observed policy event, public flight data, and independent DOT time valuations.

full rationale

The paper's derivation is self-contained in the relevant sense. The treatment is an external policy event (Selective Availability disabled on May 2, 2000), the outcome is the constructed variable adddelay = arrival delay - departure delay from public BTS data, and Equation 3 estimates a difference-in-differences coefficient beta3 on the treated x GPS interaction. The welfare calculation multiplies this coefficient by T-100 passenger counts and the DOT's 2003 value-of-travel-time estimate ($28.60/hour), all external inputs. Nothing in the chain defines the estimated effect in terms of the welfare total or vice versa: the $268 million figure is arithmetic applied to an independently estimated coefficient, not a fitted parameter relabeled as a prediction. The window-specific and distance-specific regressions revisit the same identification strategy but do not reduce to their inputs by construction; they are alternative estimates, and the acknowledged limitations (lack of a natural control, weather proxied by cancellations, possible rent capture by airlines) are identification or external-validity concerns, not circularity. Citations to Weeden and Buchholz are background or acknowledged assistance and are not load-bearing for the central result. No self-citation chain or uniqueness theorem is invoked to force the choice of model. Accordingly, no circular step was found.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper's central estimate rests on a handful of assumptions: the DID identifying assumption, the interpretation of adddelay as capturing GPS-driven in-flight delay, the immediate adoption of improved GPS by airlines, and the DOT's value of time. No free parameters are fitted by the author; the regression coefficients are estimated, not imposed.

assumptions (4)

domain assumption Parallel trends in adddelay between the control year 1999 and treated year 2000 for the seasonal component
Section 3.1: the DID relies on flights in 1999 serving as the control; any year-specific change in the seasonal delay pattern confounds the estimate.
domain assumption The difference between arrival and departure delay (adddelay) isolates in-flight delays relevant to GPS accuracy
Section 2.1: the author constructs adddelay on the expectation that GPS affects airborne navigation; other in-flight factors (weather, ATC, routing) also affect it.
domain assumption Airlines and pilots were able to benefit from improved GPS within the study window
Section 3.1: the author assumes immediate adoption and acknowledges that some planes may have lacked receivers or training.
domain assumption DOT's value of travel time ($28.60/hour in 2000 dollars) approximates passengers' marginal willingness to pay for reduced flight time
Section 3.2: welfare conversions use DOT's 2003 guidance; if the true VOT differs, the dollar estimates scale proportionally.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Valuing Diffuse Global Public Goods from Satellite Constellations: Evidence from GPS and Airline Delays." pith.science (2026). https://pith.science/paper/IMTRVTYK

@misc{pith2026250608209,
  author       = {Pith},
  title        = {Pith review of: Valuing Diffuse Global Public Goods from Satellite Constellations: Evidence from GPS and Airline Delays},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IMTRVTYK}},
  note         = {Machine review of arXiv:2506.08209}
}

abstract

This paper studies the welfare impact of discrete improvements to global public goods in the context of the Global Positioning System (GPS). Specifically, I find that by disabling Selective Availability in May, 2000, and thus significantly increasing the accuracy of GPS, the United States generated at least \$268 million (2000 dollars) of additional welfare gains. To quantify this welfare impact, I apply a difference-in-differences model to the Bureau of Transportation Statistics's Airline On-Time Performance Data in the years 1999 and 2000. I use this model to estimate the time saved per flight attributable to the improved GPS and multiply these time savings by the number of passengers in the ensuing year and their values of time. I conclude by estimating the economic loss from current threats to the provision of satellite-based global public goods.

Figures

Figures reproduced from arXiv: 2506.08209 by the authors.

**Figure 2.** Total Airline Passengers Each Month The DOT’s economists provide a path to understand how much consumers valued the opportunity cost of the time they spent on the airplane. They publish an updated assessment every few years of the valuation consumers assign to the time they spend on various forms of transportation. One such assessment was published in 2003, however their calculations were based on 2000 numbers, the … view at source ↗

**Figure 4.** Average Regression Coefficients for Flights Over and Under 2000 Miles [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: Calculated Welfare Gains Per Month The increased time saved in the early months combined with the increased efficiency of the model from subdividing flights by distance leads to the early peak seen in the graph. Two possible explanations for this peak are that pilots a…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

BTS relies on informal procedures to verify the accuracy of on-time performance data and lacks procedures to verify the data’s completeness

2025 legislative session: Budget information — washington department of fish & wildlife . (n.d.). Retrieved February 9, 2025, from https://wdfw.wa.gov/about/administration/ budget/update Airline on-time performance data. (n.d.). https://www.transtats.bts.gov/DatabaseInfo.asp? QO VQ=EFD&Yv0x=D Blatt, B. (2024). Airlines are padding flight times. it’s not y...

work page arXiv 2024
[2]

There were only 7 flights with negative air times. These negative flight times are most likely an error in reporting and due to the small number of affected flights relative to the size of the dataset, it would not affect the results of the study to remove them. Without the removed flights, the minimum arrival delay changed to -989. This value appears ill...

work page 1999

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

BTS relies on informal procedures to verify the accuracy of on-time performance data and lacks procedures to verify the data’s completeness

2025 legislative session: Budget information — washington department of fish & wildlife . (n.d.). Retrieved February 9, 2025, from https://wdfw.wa.gov/about/administration/ budget/update Airline on-time performance data. (n.d.). https://www.transtats.bts.gov/DatabaseInfo.asp? QO VQ=EFD&Yv0x=D Blatt, B. (2024). Airlines are padding flight times. it’s not y...

work page arXiv 2024

[2] [2]

There were only 7 flights with negative air times. These negative flight times are most likely an error in reporting and due to the small number of affected flights relative to the size of the dataset, it would not affect the results of the study to remove them. Without the removed flights, the minimum arrival delay changed to -989. This value appears ill...

work page 1999