As someone who frequently conducts surveys as part of my performance improvement work, I am paying special attention to polling that is being done to determine who will win each U.S. presidential primary. A poll of voters is a special form of a survey, similar to an employee engagement survey or a customer satisfaction survey. The result from any of these different forms of surveys is only an estimate of what a population believes or will do. This estimate, due to a variety of errors inherent in surveys of population samples, always falls within a range of percentages. So when a pollster reports that 39% of New Hampshire democrats surveyed said that they will vote for Obama and 31% will vote for Clinton, what they really mean is that we can be relatively confident that the actual percentage for that population will be within a range (usually plus or minus 3% to 6% depending on how many people were surveyed and the bias in who was selected for the survey). Given this characteristic of survey results, it is not surprising from a purely statistical perspective that Clinton won the New Hampshire primary, regardless of what happens with the re-count of votes.

Surveys are highly useful tools in getting a sense of the thinking of a particular population of people and in stimulating their thinking about important issues. But as with any tool we should respect its usefulness and limitations. When I survey employees about what they learned in a leadership program, how they are applying what they learned, and how the organization can help to maximize the impact of what they learned, I don’t assume that the percentage who indicated business results from the program is the actual percentage that got results. I know that number is in the middle of a range of what is likely to include the true percentage. That level of precision is usually sufficient for my purposes. I just need to remember that if I survey that population again and get another number that is a few percentage points different, it could be that, statistically speaking, there is no change.

Comment