This is an archived article and the information in the article may be outdated. Please look at the time stamp on the story to see when it was last updated.

DENVER (KDVR) — Data Desk crunched the numbers on this Groundhog Day and found Punxsutawney Phil, Pennsylvania’s groundhog weatherman, is less accurate for the Denver metro area than a coin toss.

Data Desk will walk you through its methodology. We first determined how many early springs Phil had predicted.

Phil was absent several years after he began making official yearly prognostications in 1887. In the past 125 decisive predictions he has made since then, he has overwhelmingly seen his shadow – which means six more weeks of winter. In only 19% of predictions has Phil looked toward an early spring.

Phil has not seen his shadow and predicted early springs more and more frequently during the more recent part of his career.

Of Phil’s no-shadow predictions, 50% have occurred since 1990, or in the most recent 30% of time. Within that slice of time, half of Phil’s early spring predictions have come in the last ten years.

Data Desk focused on this chunk of time since it is both varied and its data readily available.

Punxsutawney Phil’s predictive accuracy is and always has been hard to measure. The United States encompasses a massive array of weather system variations. Colorado contains the same breadth and variety. Data Desk has therefore confined the analysis of Phil’s predictive accuracy to the Denver metro area and within the most recent 15 years.

The graph below measure the number of degrees each March in which both the average low and average high temperatures were different from the most previous two-decade norm.

An “early spring” is considered any year in which the average March high temperature and low temperature were both warmer normal. There are nine such years between 2007-19: 2007, 2009, 2011, 2012, 2014, 2015, 2016, 2017 and 2018.

Data Desk then compared Phil’s prognostications against its own definitions to see which matched.

Between 2007-19, Phil’s predictions matched Data Desk’s definitions in five years: 2007, 2008, 2010, 2011 and 2016. The years where prediction and definition match are in red, while those that varied are in blue.

This gives Punxsutawney Phil a predictive accuracy for Denver of 38%. This mean he is less accurate than a coin but many degrees more accurate than dice or roulettes.