Monday, July 28, 2014

Revere tornado is first in Suffolk County since 1950

A look at the frequency of tornadoes in Massachusetts counties
by BRENT BENSON

The coastal city of Revere was hit by a rare tornado Monday morning, leaving a great deal of damage to trees, houses, and cars, but thankfully no serious injuries or deaths.

A look at tornado history data collected from the National Oceanic and Atmospheric Administration and presented by the Tornado History Project shows that, until today, not one tornado has been reported in Suffolk County, which contains Boston, Chelsea, Revere, and Winthrop. The only other Massachusetts County that has not reported a recent tornado is the island county of Nantucket.

MA tornadoes by county since 1950

Worcester County in Central Massachusetts has the highest number of tornadoes, followed by Franklin and Hampden counties. The other counties with the fewest tornadoes are the coastal and island counties of Barnstable, Dukes, and Nantucket.

MA tornadoes by county chart

Thursday, July 24, 2014

Globe article underestimates voter support for Governor Patrick's plan to house immigrant children

Support of 50% to 43% is a sizable advantage, even with 4.9% margin of error
by BRENT BENSON

The Boston Globe and SocialSphere released a poll of 404 likely voters on July 23, which showed 50% of the respondents supporting Governor Patrick's plan to house 1,000 immigrant children on Massachusetts military bases, 43% of the respondents opposing, and 8% with no opinion.

The accompanying article stated the 50% to 43% margin is "within the poll’s margin of error of plus or minus 4.9 percentage points," with implications that that this margin is not robust and the sentiment of Massachusetts's voters is not clear.

However, a closer look at the statistical implications of the 4.9% margin of error shows the chance is less than 5% that more voters oppose Governor Patrick's plan than support it.

A margin of error of 4.9% implies a standard error of 2.45%. We can visualize the probability distribution around the 50% support number using a probability density graph. The total area under the probability density curve is 1 and we can look at the area under various subsections to estimate likelihoods. There is a much higher likelihood—more area under the curve—for support percentages close to the survey result of 50%, with smaller and smaller probabilities as support levels get farther away from the measured value.

If we look at a support level of 46% (around where an even split between support and opposition might occur—a somewhat generous interpretation) we measure the area under the curve to the left and see a probability of 5% that the actual level of support is less than 46%. There is a 95% chance the level of support is above 46% and that more voters support the plan, than oppose it.

Probability density of support level
Care should be taken when interpreting sampling error in poll analysis to not over- or underestimate the uncertainty it implies. The 4.9% margin of error in the Globe/SocialSphere poll does not diminish the strong indication of voter sentiment in favor of Governor Patrick's plan.

Update (7/25/2014) - MassINC Polling President Steve Koczela noted that there is an interaction between support percentage and opposition percentage that leads to a slightly more complicated ballot lead calculation. The result is not much different, with a probability of 92% that supporters outnumber opposers.