about · email me · subscribe

spurious scholar

Because if p < 0.05, why not publish?

Step 1: Gather a bunch of data. There are 25,156 variables in my database. The data ranges from the mundane (air pollution in Chicago) to the weird (Hotdogs consumed by Nathan's Hot Dog Eating Competition Champion) to the super-niche (How clickbait-y Numberphile YouTube video titles are, as rated by an AI).
Step 2: Dredge that data to find random correlations between variables. "Dredging data" means taking one variable and correlating it against every other variable just to see what sticks. It's a dangerous way to go about analysis, because any sufficiently large dataset will yield strong correlations completely at random.

Fun fact: the chart used on the wikipedia page to demonstrate data dredging is also from me. I've been being naughty with data since 2014.

Step 3: Calculate the correlation coefficient, confidence interval, and p-value to see if the connection is statistically significant. "Statistically significant" is a misleading term. It sounds like it means "statistically significant" because, you know, those are the same two words. Unfortunately statistical significance is a technical term that means mumble mumble at least as extreme mumble mumble null hypothesis mumble mumble probability mumble mumble p-values.

You know what? Forget the technical definition. "Statistically significant" just means "someone did some fancy math."

I really did the fancy math below and you can check it by clicking on the "view detailed data" link under each paper. And these really do qualify as "statistically significant" in the technical sense. It's just that "statistically significant" does not mean the results are "significant."

Step 4: If it is, have a large language model draft a research paper.
Step 5: Remind everyone that these papers are AI-generated and are not real. Seriously, just pick one and read the lit review section. The silliness of the papers is an artifact of me (1) having fun and (2) acknowledging that realistic-looking AI-generated noise is a real concern for academic research (peer reviews in particular).

The papers could sound more realistic than they do, but I intentionally prompted the model to write papers that look real but sound silly.

Also: every page says "This paper is AI-generated" at the bottom and the first letters of the names of the authors always spell out C-H-A-T-G-P-T.

Step 6: ...publish:

The Tennessee Tulip: Tracing the Ties between Tenacious Troops of Tennessean Judicial Law Clerks and the Triumphs of Mizuho Financial Group's Stock Price
The Journal of Financial Flora and Fauna
r=0.954 · 95% conf. int. [0.829,0.988] · r2=0.911 · p < 0.01
Generated Jan 2024 · View data details

Genetically Modified Ouch: A Study on the Relationship Between GMO Cotton in Louisiana and the BNS Stock Price
The Journal of Agricultural Mutations and Financial Impact
r=0.888 · 95% conf. int. [0.733,0.955] · r2=0.788 · p < 0.01
Generated Jan 2024 · View data details

Whisking Degrees: The Flourishing Relationship Between Family and Consumer Sciences Associate Degrees and NYSE Composite Index Annual Percentage Change
The Journal of Culinary Finance and Consumer Studies
r=0.604 · 95% conf. int. [0.006,0.884] · r2=0.365 · p < 0.05
Generated Jan 2024 · View data details

The Tantalizing Tango: Tracing the Ties between Tunes and Turbines - A Goofy Good-Hearted Glance at the Association of Associates Degrees in Music and Dance with Solar Power Generation in Costa Rica
The Journal of Eclectic Energy and Euphonious Endeavors
r=0.988 · 95% conf. int. [0.952,0.997] · r2=0.976 · p < 0.01
Generated Jan 2024 · View data details

Engineering Technologies: Energizing Cameroon's Renewable Energy Sector
The Journal of Renewable Energy Innovations
r=0.961 · 95% conf. int. [0.838,0.991] · r2=0.923 · p < 0.01
Generated Jan 2024 · View data details

Sparking Interest: A Shocking Connection Between Bachelor's Degrees in Biological and Biomedical Sciences and Electricity Generation in Tanzania
The International Journal of Bioelectricity and Ecological Innovations
r=0.995 · 95% conf. int. [0.980,0.999] · r2=0.991 · p < 0.01
Generated Jan 2024 · View data details

The Bar Examined: A Statistical Analysis of the Correlation Between 7th Grade Public School Enrollment and Lawyer Population in the United States
The Journal of Legal Demographics
r=0.844 · 95% conf. int. [0.704,0.920] · r2=0.712 · p < 0.01
Generated Jan 2024 · View data details

Cotton Genetically Modified to Make Us Say 'I Can't Even': An Unlikely Correlation Analysis
The Journal of Quirky Genetic Modifications
r=0.907 · 95% conf. int. [0.771,0.964] · r2=0.823 · p < 0.01
Generated Jan 2024 · View data details

Sprouting Connections: Exploring the Relationship Between GMO Soybeans in South Dakota and LPG Consumption in Hong Kong
The Journal of Ecological Interconnections
r=0.916 · 95% conf. int. [0.805,0.965] · r2=0.838 · p < 0.01
Generated Jan 2024 · View data details

GMO Growing in Indiana: Gauging the Gain in Rogers Communications' Stock Price
The Journal of Agricultural Finance and Telecommunications Economics
r=0.926 · 95% conf. int. [0.823,0.970] · r2=0.857 · p < 0.01
Generated Jan 2024 · View data details

Cottoning On to Romance: The Genetically Modified Connection Between Tennessee Cotton and xkcd Comics
The Journal of Agricultural Genetics and Pop Culture Neuroscience
r=0.914 · 95% conf. int. [0.755,0.971] · r2=0.835 · p < 0.01
Generated Jan 2024 · View data details

The Corny Connection: Exploring the GMO Influence on the Legal Landscape
The Journal of Genetic Modification and Legal Implications
r=0.966 · 95% conf. int. [0.909,0.988] · r2=0.933 · p < 0.01
Generated Jan 2024 · View data details

Aubrey's Appellation Association: Analyzing the Allure of Aubrey on Justin Upton's Annual Athletic Achievements
The Journal of Sports Psychology and Player Preferences
r=0.836 · 95% conf. int. [0.581,0.941] · r2=0.699 · p < 0.01
Generated Jan 2024 · View data details

Kicking Off the Event Horizon: Exploring the Goalpost Connection Between FA Cup Final Goal Difference and Google Searches for 'Black Hole Photo'
The Journal of Cosmic Coincidences
r=0.794 · 95% conf. int. [0.431,0.936] · r2=0.630 · p < 0.01
Generated Jan 2024 · View data details

Cultured Coincidence: The Yogurt-Scoring Nexus in the New England Patriots' NFL Seasons
Journal of Probiotic Performance
r=0.791 · 95% conf. int. [0.610,0.893] · r2=0.625 · p < 0.01
Generated Jan 2024 · View data details

Connecting Communications Credentials and Alabama's Ample Amperage: A Alluring Analysis
The Journal of Electrifying Linguistics
r=0.905 · 95% conf. int. [0.641,0.978] · r2=0.819 · p < 0.01
Generated Jan 2024 · View data details

Bartering Business Badges: A Bizarre Beeline Between Business Management Bachelors and Barrels of Benzene in Guam
The Journal of Unconventional Commerce and Eclectic Exchanges
r=0.980 · 95% conf. int. [0.924,0.995] · r2=0.961 · p < 0.01
Generated Jan 2024 · View data details

Business Bachelor's and Booming Bell: Exploring the Link between Business Degrees and AT&T Customer Satisfaction
Journal of Business and Telecommunications Research
r=0.962 · 95% conf. int. [0.844,0.991] · r2=0.926 · p < 0.01
Generated Jan 2024 · View data details

Playing with Fire: Arson in Minnesota and xkcd Comics on Shimmering Romance
Journal of Pyromania Studies
r=0.923 · 95% conf. int. [0.788,0.973] · r2=0.852 · p < 0.01
Generated Jan 2024 · View data details

On the Rocks: The Margarita Name's Connection to Motor Vehicle Thefts in Indiana
Journal of Quirky Sociological Studies
r=0.956 · 95% conf. int. [0.916,0.977] · r2=0.914 · p < 0.01
Generated Jan 2024 · View data details

Milk Chug-A-Lug and Burglary Huggerie: An Udderly Surprising Connection in Michigan
The Journal of Dairy Dynamics
r=0.968 · 95% conf. int. [0.935,0.984] · r2=0.937 · p < 0.01
Generated Jan 2024 · View data details

Cheese and Crime: Unraveling the Curd-ious Connection Between Cottage Cheese Consumption and Burglaries in Massachusetts
The Journal of Dairy Delinquency
r=0.919 · 95% conf. int. [0.839,0.960] · r2=0.844 · p < 0.01
Generated Jan 2024 · View data details

The Celestial Burglary: Unveiling the Unconventional Relationship Between Neptune's Distance and Burglaries in Wyoming
The Journal of Interplanetary Criminology
r=0.973 · 95% conf. int. [0.948,0.986] · r2=0.946 · p < 0.01
Generated Jan 2024 · View data details

Heist in the High Plains: A Study on the Unlikely Relationship between Robberies in Montana and US Hospital Occupancy Rate
The Journal of Unusual Correlations in Social Science
r=0.747 · 95% conf. int. [0.431,0.900] · r2=0.558 · p < 0.01
Generated Jan 2024 · View data details

Spreading the Wealth: A Churn for the Butter - AvalonBay Communities' (AVB) Stock Price Response to Butter Consumption
The Journal of Gastronomical Economics
r=0.885 · 95% conf. int. [0.727,0.954] · r2=0.783 · p < 0.01
Generated Jan 2024 · View data details


Currently viewing 25 of 4,731 spurious research papers

Page
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190



Why this works

  1. Data dredging: I have 25,156 variables in my database. I compare all these variables against each other to find ones that randomly match up. That's 632,824,336 correlation calculations! This is called “data dredging.” Fun fact: the chart used on the wikipedia page to demonstrate data dredging is also from me. I've been being naughty with data since 2014.
    Instead of starting with a hypothesis and testing it, I isntead tossed a bunch of data in a blender to see what correlations would shake out. It’s a dangerous way to go about analysis, because any sufficiently large dataset will yield strong correlations completely at random.
  2. Lack of causal connection: There is probably no direct connection between these variables, despite what the AI says above. Because these pages are automatically generated, it's possible that the two variables you are viewing are in fact causually related. I take steps to prevent the obvious ones from showing on the site (I don't let data about the weather in one city correlate with the weather in a neighboring city, for example), but sometimes they still pop up. If they are related, cool! You found a loophole.
    This is exacerbated by the fact that I used "Years" as the base variable. Lots of things happen in a year that are not related to each other! Most studies would use something like "one person" in stead of "one year" to be the "thing" studied.
  3. Observations not independent: For many variables, sequential years are not independent of each other. You will often see trend-lines form. If a population of people is continuously doing something every day, there is no reason to think they would suddenly change how they are doing that thing on January 1. A naive p-value calculation does not take this into account. You will calculate a lower chance of "randomly" achieving the result than represents reality.

    To be more specific: p-value tests are probability values, where you are calculating the probability of achieving a result at least as extreme as you found completely by chance. When calculating a p-value, you need to assert how many "degrees of freedom" your variable has. I count each year (minus one) as a "degree of freedom," but this is misleading for continuous variables.

    This kind of thing can creep up on you pretty easily when using p-values, which is why it's best to take it as "one of many" inputs that help you assess the results of your analysis.
  4. Outliers: Some datasets here have outliers which drag up the correlation. In concept, "outlier" just means "way different than the rest of your dataset." When calculating a correlation like this, they are particularly impactful because a single outlier can substantially increase your correlation.

    Because this page is automatically generated, I don't know whether any of the charts displayed on it have outliers. I'm just a footnote. ¯\_(ツ)_/¯
    I intentionally mishandeled outliers, which makes the correlation look extra strong.



Spurious Scholar was launched January 27, 2024. If you have feedback on it, I'd love to hear from you! Shoot me a note: feedback@tylervigen.com.


Project by Tyler Vigen
emailme@tylervigen.com · about · subscribe


CC BY 4.0