Keeping our Polling Data Real

Oct 7, 2016 | Press

By: Eli Ackerman, Director, Polling & Partner Ops

Readers may have noticed that the Presidential tracker did not update Thursday, 10/6. If it had, the results would have shown a big jump in support for Clinton from an 8 point lead after Tuesday to a 12 point lead after Wednesday. Because there was not a news event that would have precipitated such a massive gain for Clinton, we dug further into our data and found reason to believe the data collected on Wednesday represented a significant outlier.

In particular, the results among senior citizens were perplexing. Among senior citizens, Clinton beat Trump by 15 points. That result was simply not close to being in line with data that Lucid had collected historically. While Clinton has gained substantial ground with seniors since her first debate with Donald Trump, she and Trump are running neck-and-neck among that demographic.

Looking at our daily results among seniors, just for the last week you can see how much the 10/5 result stands out both as an unusual high for Clinton and also a low in terms of the total number of Americans aged 65+ interviewed.

Pasted image at 2016_10_07 12_50 PM

This next chart shows a rolling 7-day average among senior citizens dating back three weeks to 9/22. Prior to the debate on 9/26, Clinton clearly trailed Trump among seniors by between 4 and 8 points and began gaining ground immediately after the debate, drawing even as of late last week.

Pasted image at 2016_10_07 12_50 PM (1)

Accordingly, the single day numbers for Clinton on 10/5 were not tuned to historic trend lines .

After ensuring nothing changed in its methodology on that day in particular, we increased the number of surveys conducted yesterday to see if a larger dataset would confirm movement among seniors or confirm that the day prior had been an outlier. In fact, among seniors interviewed yesterday, Clinton and Trump ran dead-even: 43% to 43%.

We’ve learned a couple of lessons here. The first is the inherent variability of data on a day-to-day basis with smaller N sizes. Going forward, we’ll be aiming to increase the daily number of surveys we conduct to mitigate the risk for future outliers. The second lesson is that data collection is inherently challenging. While results are often discussed as scientific, methodological choices along the way will have an impact on results.

We’re not afraid to share those choices and challenges with you along the way and welcome your advice, questions, and constructive criticism.

Recommended Posts

Stay in the know with LUCID. Subscribe to our newsletter.