Project 05: Mercury Retrograde — Real Effect or Perception Bias? #

Source: bigastrologybook.com/2/research/19/project-5 Archive Date: 2026-03-21 Book: The Big Astrology Book of Research by Renay Oshop Scope: 24 years of daily travel/incident data (2001–2024); Bayesian NLP framework; ~72 Mercury retrograde periods

Research Question #

Does Mercury retrograde actually correlate with increased technology failures, travel disruptions, and communication breakdowns — or is the widespread experience of retrograde chaos a product of perception bias?

Hypothesis #

The Mercury retrograde effect is primarily a psychological phenomenon: confirmation bias and calendar confounds (holiday seasons, weekly patterns, annual travel volume) account for the perceived correlation between Mercury's apparent backward motion and terrestrial disruption.

A Critical Upfront Disclosure #

The travel and incident data in this study is synthetic. It was generated algorithmically to model realistic seasonal patterns, day-of-week effects, holiday spikes, and long-term trends — based on Bureau of Transportation Statistics averages — but it is not actual flight delay or technology outage data. Similarly, the social media corpus analyzed for attribution bias consists of 20,000 simulated posts, not real tweets or Reddit comments.

What is real: Mercury's retrograde periods, calculated precisely via Swiss Ephemeris (accurate to <0.01°). The astronomical timeline is exact; the incidents it is tested against are illustrative.

This matters for how the results should be read: the findings demonstrate what a rigorous methodology would find, and establish a framework for testing the hypothesis with real data. They do not constitute empirical evidence about actual Mercury retrograde effects on the real world. The methodology is the contribution.

Why the Question Is Harder Than It Looks #

Mercury retrograde is one of astrology's most culturally pervasive claims, and one of the most testable — in principle. The problem is that naive tests almost always find spurious correlations, not because Mercury retrograde is real, but because of calendar confounds that track with retrograde periods:

• Holiday overlap: Mercury is in retrograde roughly three times per year, each period lasting three weeks. Late November and late December retrograde periods overlap with Thanksgiving and Christmas — the two most travel-disrupted times of year regardless of any planetary influence.
• Summer travel: Retrograde periods in June–July coincide with peak travel season, when delays are highest for purely terrestrial reasons.
• Weekly rhythms: Friday and Sunday show elevated incident rates year-round. If a retrograde period contains more Fridays, it will look more incident-prone than it is.

Anyone who has lived through a "chaotic" Mercury retrograde period and later noticed it was also a holiday season, or a busy travel week, is experiencing exactly this confound. The detrending methodology here is designed to remove these layers before any comparison is made.

Data #

Total retrograde periods: ~3 per year × 24 years = ~72 periods, totaling 1,686 retrograde days out of 8,766 total (19.2% of the time).

Methods #

1. Multiplicative Time-Series Decomposition #

The core analytical tool. Before any Mercury comparison is made, the raw daily incident count is decomposed to remove known confounds:

Anomaly Ratio = Observed / (Trend × Season × Day-of-Week × Holiday)

• Trend: Long-term growth in air travel and internet usage across 24 years
• Season: Annual cycle — summer peak, winter lull
• Day-of-week: Friday/Sunday travel peaks; Monday tech support spikes
• Holiday: Explicit adjustment for the ~10 highest-impact US holidays

What remains is the Anomaly Ratio: how much the day deviates from what the calendar alone would predict. An Anomaly Ratio of 1.00 means exactly as expected. Only this residual is compared to Mercury's status.

2. Bayesian Inference #

Rather than a single p-value, the analysis uses a Bayesian framework to continuously update belief in the "retrograde hypothesis" as each day's data is observed. The prior is agnostic (50/50). The Bayes Factor compares how well the "retrograde causes incidents" hypothesis predicts the data versus the simpler null model ("incidents are random").

3. NLP Attribution Analysis #

The synthetic social media corpus distinguishes between:

• Objective complaints: "WiFi is down," "flight cancelled" — generic frustration
• Astrological attributions: "Mercury is retrograde" — explicit causal assignment to planetary motion

This separation is what makes the bias analysis meaningful. It's not enough to count complaints during retrograde; you need to know whether people are blaming the planet specifically, and whether that blame rate tracks with actual incident rates or diverges from them.

Results #

1. The Detrending Result: Null #

After removing the holiday, seasonal, day-of-week, and trend layers:

T-test: p = 0.21 — not significant.

The difference between retrograde and direct periods in detrended data is 0.25%. This is noise. Once the calendar is properly accounted for — once you remove the Christmas travel chaos, the summer delays, the weekend spikes — Mercury retrograde adds nothing to the model.

This is the finding that matters: the raw data almost certainly shows elevated incidents during retrograde periods, but that elevation is fully explained by when retrograde periods fall in the calendar, not by Mercury's motion itself.

2. The Bayesian Result: The Hypothesis Loses Ground #

The Bayes Factor of 0.83 means the data provides mild evidence against the retrograde hypothesis — not strong enough to call it disproven, but in the wrong direction entirely for the hypothesis to survive. A rational agent observing this data should update their belief in Mercury retrograde's causal role slightly downward, from 50% to 45%.

The posterior decline is modest because the data is noisy (as any 24-year incident dataset will be). With real data and a larger corpus, the update would likely be more decisive in either direction.

3. The Confirmation Bias Mechanism #

Why does the perception of Mercury retrograde chaos persist despite the absence of a real signal? The analysis points to two well-understood cognitive mechanisms:

Selection bias: People are more alert to disruptions during retrograde periods because they've been primed to notice them. The same WiFi outage that goes unremarked on a Tuesday in April becomes "classic Mercury retrograde" in November. This selective encoding creates a false signal in personal memory.

Attribution error through calendar confound: The observer fails to detrend their own experience. Late December travel chaos feels like Mercury retrograde chaos because it coincides with Mercury retrograde — but it would occur regardless. The astrological attribution is post-hoc rationalization of what the calendar was going to produce anyway.

The analysis frames this precisely: "The fault, dear Brutus, is not in our stars, but in our calendars." Once you know when Mercury retrogrades, you can predict which events will be blamed on it — not because Mercury caused them, but because retrograde periods reliably fall during the most disruption-prone times of year.

What Would Make This Study Definitive #

The framework here is sound; the data is not. To move from methodological demonstration to empirical conclusion, this study needs:

1. Real incident data: FAA/BTS flight delay records, Downdetector tech outage logs, or insurance claim databases — any large, timestamped incident dataset that doesn't know about Mercury's position
2. Real social media data: Actual Twitter/Reddit posts with timestamps, enabling authentic attribution rate analysis rather than simulated posts
3. Pre-registration: The detrending methodology should be specified before the analysis to prevent unconscious parameter tuning

Applied to real data with this methodology, the result would either confirm the null (most likely, given the analytical framework) or reveal a genuine residual signal that survived rigorous detrending — which would itself be extraordinary and worth pursuing seriously.

Conclusion #

After mathematically removing the calendar confounds that plague naive retrograde studies — holiday spikes, seasonal travel variation, weekly rhythms, long-term trends — Mercury retrograde's association with travel and technology disruption vanishes. The detrended anomaly ratio during retrograde periods (1.002) is statistically indistinguishable from direct periods (0.9995). The Bayesian posterior for the retrograde hypothesis decreases on exposure to the data.

The more important finding is why the perception persists despite the absence of a real signal: Mercury retrograde reliably falls during the most calendar-disrupted times of year, and human memory is predisposed to notice and remember disruptions that have a ready explanation. The planet is blameless. The calendar is the culprit.

The rigorous test remains to be done with real incident data. The methodology to do it correctly is documented here.

Archived code and synthetic data outputs preserved in backup/. Real Mercury retrograde date calculations are reproducible via Swiss Ephemeris.

Mercury Retrograde Perception Bias #

Hypothesis #

"Mercury Retrograde causes technology failures, travel delays, and communication breakdowns."

This project challenges the popular astrological claim by distinguishing between Actual Events (Signal) and Perception Bias (Noise). We employ a high-power statistical approach to determine if Retrograde periods correlate with objective failures, or if the phenomenon is primarily a psychological confirmation bias.

Methodology: "Hard Science" Upgrade #

Moving beyond simple visual checks, this analysis implements the rigorous standards established in our Biological Rhythms study (Project 3):

1. Multiplicative Time-Series Decomposition (Detrending)

   • We isolate "True Anomalies" by mathematically removing known cycles:
     • Weekly Cycle: Friday/Sunday travel peaks.
     • Seasonal Cycle: Summer travel volume vs Winter lulls.
     • Yearly Trend: Long-term growth in air travel/internet usage.
     • Holiday Factors: Explicit adjustment for Thanksgiving, Christmas, and July 4th spikes.
   • Formula: $Residual = Observed / (Trend \times Season \times Holiday \times DayOfWeek)$

2. Natural Language Processing (NLP)

   • We analyze a corpus of "User Reports" (simulated social media stream) to detect Attribution Bias.
   • We categorize posts into "Objective Complaints" (e.g., "Wifi down") vs "Astrological Attributions" (e.g., "Mercury is in retrograde!").

3. Bayesian Inference

   • We calculate the posterior probability of the Retrograde Hypothesis ($H_1$) given the observed data.
   • $P(H_1 | Data) = \frac{P(Data | H_1) \cdot P(H_1)}{P(Data)}$
   • We continually update our belief in the "Retrograde Effect" as new data points (days) are observed.

Data Sources #

1. Travel & Logistics:
   • Simulated Daily Flight Incident dataset (N=8,766 days, 2001-2024).
   • Modeled on Bureau of Transportation Statistics (BTS) monthly averages + Holiday logic.
2. Astronomical Timeline:
   • Daily Mercury Status (Retrograde/Direct) calculated via Swiss Ephemeris.
3. Sentiment Corpus:
   • N=20,000 simulated social media posts for NLP testing.

Execution #

Run the rigorous analysis:

python3 clean_analysis_v2.py

This generates:

• retrograde_residuals.png: KDE plot of detrended residuals.
• analysis_results_clean.csv: Full daily dataset with Mercury status and Anomaly Ratios.

Data Provenance #

Flight Delay Data #

• Source: Bureau of Transportation Statistics (BTS).
• Dataset: Airline On-Time Performance Data.
• Link: https://www.transtats.bts.gov/OT_Delay/OT_DelayCause1.asp