The 2012 American Community Survey Data show numerical changes in the way Arlingtonians get to work, but the differences are not statistically significant.
Yesterday, the Census Bureau released new American Community Survey (ACS) data for 2012. Between 2011 and 2012, Arlington residents maintained their high rates of biking, walking, and transit use, with small percentage changes that disappear when we include tests of statistical significance.
The change with the strongest (albeit, still weak) change, statistically speaking, was the increase in teleworking or working from home, which is in line with the Transportation Planning Board’s recent initial findings from the 2013 State of the Commute survey. The second-strongest (though also weak) change was the drop in transit as a share of commute modes. While Arlington County offers multiple transit services, these figures are consistent with observations that Metrorail ridership declined between 2011 and 2012, and have continued to decline into 2013.
Comparing Arlington Residents’ Commute Mode Split from 2011 and 2012
Sources: American Community Survey 1-Year Estimates, 2011 and 2012
When comparing ACS figures over time, (and reading news reports about trends based on the ACS) it is important to remember that the Survey’s results are a sample, not a census. In other words, the Census Bureau does not ask all Arlington residents about their commutes every year; rather, they ask a few people, and then apply the answers that they get to all Arlingtonians.
Sampling is necessary because conducting a census is very expensive, but sampling does not give us a completely accurate picture of the population, because your numbers might change depending on who responds to your survey. So when we compare figures over time, we need to be sure that two figures are different enough to account for an actual change in the population’s behavior, not just because the surveys happened to ask two different groups of people.
Given the margins of error reported by the Census Bureau for the 2011 and 2012 data, we get z-scores of less than 1.645. The z-scores would need to be larger than 1.645 in order to meet the lowest confidence level (90%) usually accepted in tests of difference. For more information on the ACS and statistical analysis of the Survey results, see A Compass for Understanding and Using American Community Survey Data: What General Data Users Need to Know (PDF, 1.6MB).
Correction: An earlier version of this post incorrectly identified the organization that produced the 2013 State of the Commute survey as the Transportation Research Board instead of the Metropolitan Washington Council of Government’s Transportation Planning Board. The error has been corrected.