Polling

I had planned to write more about Apple and AI this week, but another topic keeps pinging my brain. Polling. Political polling. No matter what source of news you might read, it is laden with poll information. Are they accurate, I wonder? How do they reach people? My goal in this post is to shed some light on the topic.

Political polls have become a mainstay when it comes to elections, in theory offering a snapshot of public opinion. Political polling has its roots in the early 20th century, with the invention of the telephone playing a pivotal role. Early polls relied 100% on landlines, reaching a broad swathe of the population. However, the rise of cell phones and declining landline usage has forced pollsters to adapt.

A crucial aspect of polling is ensuring a representative sample. Pollsters typically use a technique called Random Digit Dialing (RDD). This involves generating random phone numbers within a specific area code, ensuring everyone has a chance of being included. RDD picks an area code and then a random three-digit exchange and then a set of random 4-digits. Traditionally, this focused on landlines.

Then there is the cell phone conundrum. With the decline of landlines, including cell phones in polls became essential. However, it presents challenges. Cell phone numbers are often unlisted, making RDD difficult and less effective. Landline users tend to be older, more likely to live in rural areas, and skew Republican. Cell phone users are younger, more urban, and lean Democratic. This can lead to discrepancies in overall results, particularly in close races. Additionally, some people, me included, are less likely to answer calls from unknown numbers. Newer methods like mixed mode surveys (phone and online) or address-based sampling focusing on specific households may eventually help reach a broader audience.

Reaching cell phones can be more expensive for pollsters due to additional verification steps and the prevalence of pre-paid plans that limit incoming calls. Finally, as landline use continues to decline, reaching a representative sample necessitates including cell phones. Ignoring this segment would lead to increasingly inaccurate polls. New technologies like Automatic Phone Dialers (APDs) that can handle both landlines and cell phones are improving efficiency and data collection but are far from perfect.

The inclusion of cell phones in political polling has been a hot topic for over a decade. While research suggests minimal impact on overall results, there are nuances to consider. While landlines and cell phones have their pros and cons, a well-designed poll incorporating both can provide a more accurate picture of public opinion.

Polling accuracy is a constant debate. Factors like sample size, weighting demographics, and undecided voters can influence results. Polling is inherently a snapshot in time, and unexpected events can sway voter sentiment. From a statistical point of view, it is important to look at the margin of error (MOE). There are two parts to the MOE. What we see with most all polls is a plus or minus X percentage points. The other part is the confidence level the MOE is correct. A 95% confidence level means if pollsters fielded the same survey 100 times, we could expect the result to be within X percentage points of the true population value 95 of those times.

For a more specific example, suppose candidate A got 48% versus candidate B who got 43%. If the MOE with 95% confidence was +/- 3%, that means candidate A’s true poll could have been 45% to 51%. Candidate B (the underdog in the poll) could have been between 40% and 46%. In other words, the underdog could have come out on top: 46% vs. 45%. I learned a lot about statistics in research in my work on congestive heart failure in 2014. The subject of statistical accuracy for research, and for polling, is quite deep. I hope my simple example makes sense.

The polling industry is constantly evolving. Online surveys and mixed methods (combining phone and online) are gaining traction. The downside of the online approach is many of them are mainly trying to get your email address and donations. As technology changes, pollsters will need to adapt to maintain accurate and representative samples. Political polls offer valuable insights, but it’s important to understand their limitations.

I would suggest to always consider the margin of error. It reflects the potential for the polls’ results to differ from the actual population. Secondly, I would say don’t rely too much on polls. By combining polling data with other forms of political analysis, you can gain a more holistic understanding of the political landscape and avoid being misled by headlines.

Epilogue: An alternative approach to polling could be the widespread use of blockchain voting. This could enable frequent “elections” on specific topics. It does not take a large population to gain a statistically valid view of how people feel. Unfortunately, as I wrote in Election Attitude – How Internet Voting Leads to a Stronger Democracy, the trend in the United States is to make it harder to vote, not easier. In Wyoming, an article just appeared titled Hand count ballots if you don’t mind mistakes, wasting time and money. It was about a proposition being support by the Attorney General to ban use of voting machines (not connected to the Internet) to be banned. Hand count only. Rice University conducted an experiment on hand counting vs. machine counting. They provided 100 ballots and had them counted by hand multiple times. The average accuracy was 58%. The machine count accuracy was 100%.

Note: I use Gemini AI and other AI chatbots as my research assistants. AI can boost productivity for anyone who creates content. Sometimes I get incorrect data from AI, and when something looks suspicious, I dig deeper. Sometimes the data varies by sources where AI finds it. I take responsibility for my posts and if anyone spots an error, I will appreciate knowing it, and will correct it.