In 1824,

The Harrisburg Pennsylvanian, a newspaper from a town in Pennsylvania conducted the first known public opinion polls in history, and successfully predicted the result of the vote in the close race between Andrew Jackson and John Quincy. However, opinion polls do not always reflect the opinions of the whole electorate accurately, especially with limited sample sizes and time gaps between polls and actual voting. In 2016, many media outlets failed to predict the results of the Brexit referendum or the US presidential election accurately, for example. Below, poll tracker shows how poll results were misleading, especially in the tight Brexit race.

Since opinion polls have to be conducted with limited sample sizes, there are three key variables that could influence their accuracy.

1 - Sampling accuracy

The distribution of interviewees’ demographics such as location, educational level, gender, age, or religion should resemble the actual diversity of the population, and the sample size should be large enough to increase this accuracy. Like tasting a stew during cooking, as we stir better, all ingredients are mixed evenly, and we can accurately get the right taste of the stew from a small sample size.

2 - Interviewers’ bias

Because people want to avoid confrontations or want to look good in front of others, interviewees may respond with answers that may sound desirable socially or to interviewers. Interviewers may be able to push interviewees to answer certain ways because of how they ask, push polls.

3 - Time

People in general react more to recent occurrences. Electoral polls usually swing drastically with the latest scandals involving candidates, just as movies or music recently released tend to be selected for awards.

When we encounter poll results, it is important to understand these variables more than which side is leading. The context gives us a slightly clearer picture of why one side is leading, and the likelihood of seeing the opposite result. In this post, I collected many visualizations that give context to opinion polls or the vote results of the Brexit referendum and US presidential elections in 2016.

Opinions by Regions

Regions are the first layer representing demographic groups

National elections may be seen as a battle between each region, and their results reflect their different economic, cultural, and demographic backgrounds. Like the New York Times example below, many regions in England and Wales voted to exit, while Scotland and Northern Island, which are far away from the central government voted against. Greater London also voted to remain, reflecting its population consisting of more people with higher education, higher earners, and more foreign-born people, which I’ll discuss more later in the section about demographic distributions.

Geographically inaccurate but electorally accurate maps

The chart below by the Guardian is a more accurate representation of the map by distorting it with the population size of each area. In comparison to the New York Times’ map above, the presence of Greater London in voting counts is clearer.

In the US presidential elections, results in each state are the most important factor since people vote for each state’s representing party, and the number of their seats depends on the population size of each state. Therefore, distorted US maps that represent population size of each state are used often in the US election, instead of ones with geographic accuracy. Many of these maps represent each state using a square with its area representing their electoral vote counts, but it still gives a sense of the geographical relationship between each state (Data Visualization Infographics v.s. Products).

Opinions by Each Demographic Group

The following two charts by the Guardian and Financial Times both try to address trends for the Brexit votes based on demographics. For example, younger voters, degree holders, and higher earners overwhelmingly voted to remain, but younger voters consisted of a small percentage to the overall vote counts, resulting in a smaller impact to the result.

Tracking Opinions Over Time

Changes in supporting parties by demographic groups

The New York Times captured the shift in supporting parties from 2004 by household income levels. Additionally, although lower income groups supported the democratic party historically, education levels correlated even more to shifts in a supporting party for the 2016 election. The Times tracked these shifts by various demographic attributes, such as ethnicity, gender, and education.

Demographic pattern and changes in results

The Economist’s team used the data from YouGov to illustrate how supporters for each party repositioned their opinions since 2017 based on May’s deal proposed in 2019.

Plots of swing counties and their degrees

The Wall Street Journal plotted each county by their velocity of support to a party and how much that shifted from 2012. The chart below illustrates how the 2016 presidential election caused many swing counties.

Supporting parties for each state historically

Swing states, which don’t have a huge gap between their active supporters for either party, are significant in predicting the US presidential elections. The below chart by the Wall Street Journal illustrates the historical differences between each party’s supporters for each state. This chart can inform us of the trend of the support for party and political directions that are significant for each state.

Mapping various poll results over time

Voters’ opinions don’t only change between each election — they may change drastically within a single campaign period. Similar to the opinion poll tracker by the Telegraph from the top of this entry, the Guardian mapped various poll results for the 2015 UK election, and drew median lines out of these plots to estimate positions for each party throughout the campaign.

How each county swung in 2018 from 2016

The Guardian used a shifting arrow map to illustrate Democrats’ gain for the House of Representatives by showing how each region swung for the 2018 midterm since the last election. Democrats increased their support in large areas where Donald Trump dominated in 2016. Larger blue arrows demonstrate greater regrets in 2016, and tracking these velocities similarly by polls could suggest the next electoral results.

Chance of Misleading Poll Results

In 1948, Gallup predicted that 49.5% of the public would vote Thomas Dewey for the presidency, but the real result was almost the reverse: Truman for 49.6% and Dewey for 45.1%. Chicago Daily Tribune published the famous headline “Dewey Defeats Truman” based on the polling data. Although Gallup mentions that their accuracy improved dramatically after the 60s, they were wrong in recent elections including 2000, 2012, and 2016.

Nowadays, voters are online and have closer access to information including numbers of opinion polls conducted by various media outlets. If poll results suggest your supporting side was going to win, these poll results may discourage you to bother going to vote. Closer access to data is also true and crucial for candidates — last minute stories drastically influence voters’ minds, and this velocity is getting greater as modern campaigns become more online with the greater access to real-time data.

Illustrating the likelihood to swing participants’ opinions

The Economist’s team used a “ternary” plot instead of the common two-dimentional plot for the Brexit poll data — their attempt was to portray how poll participants were likely to position for supporting remain, leave, or leave without the deal based on their responses and their likelihood to change their opinions based on their demographics.

Elections also do not always represent the public opinion accurately

The Economist’s analytical team ran a model based on various polls to suggest how the 2016 US election would have resulted if all Americans voted. The simulation suggests Clinton would win over Trump, which was also suggested by predictions based on polls before the actual election.

在1824年，

宾夕法尼亚州一个小镇的报纸《哈里斯堡宾夕法尼亚州》进行了历史上首次已知的民意测验，并成功预测了安德鲁·杰克逊和约翰·昆西之间亲密比赛的投票结果。 但是，民意调查并不总是能准确反映出全体选民的意见，特别是在样本量有限以及民意调查与实际投票之间的时间间隔有限的情况下。 例如，2016年，许多媒体未能准确预测英国退欧公投或美国总统大选的结果。 下面，民意调查跟踪器显示民意调查结果如何产生误导，尤其是在激烈的英国退欧竞赛中。

由于民意测验必须以有限的样本量进行，因此存在三个可能影响其准确性的关键变量。

1-采样精度

受访者的人口统计信息(例如位置，教育水平，性别，年龄或宗教信仰)的分布应类似于人口的实际多样性，并且样本量应足够大以提高准确性。 就像在烹饪过程中品尝炖肉一样，随着我们更好地搅拌，所有成分均被混合均匀，并且我们可以从少量样品中准确地获得炖菜的正确口味。

2- 观众的偏见

因为人们想要避免对抗或想在别人面前看起来很好，所以受访者可能会做出听起来可能是社会上或受访者希望的答案。 采访者可能会因为他们的询问方式， 推动民意测验而促使受访者回答某些问题。

3-时间

人们通常对最近发生的事情有更多的React。 选举通常与涉及候选人的最新丑闻大相径庭，就像最近发布的电影或音乐往往被选为奖项一样。

当我们遇到民意测验结果时，重要的是要了解这些变量，而不是领先于哪一方。 通过上下文，我们可以更清楚地了解到一侧为何领先以及看到相反结果的可能性。 在这篇文章中，我收集了许多可视化内容，这些内容为民意调查或英国退欧公投和2016年美国总统选举的投票结果提供了背景信息。

各地区意见

区域是代表人口群体的第一层

全国大选可以看作是每个地区之间的斗争，其选举结果反映了不同的经济，文化和人口背景。 就像下面的《纽约时报》的例子一样，英格兰和威尔士的许多地区投票退出，而远离中央政府的苏格兰和北岛投票反对。 大伦敦地区也投票决定保留，以反映其人口，其中包括更多受过高等教育的人，收入更高的人以及更多在外国出生的人，我将在后面有关人口分布的部分中讨论更多。

地理上不准确但选举上准确的地图

《卫报》下方的图表通过将每个区域的人口规模进行扭曲来更准确地表示地图。 与上面的《纽约时报》的地图相比，大伦敦的投票人数更加清楚。

在美国总统选举中，自从人们投票支持每个州的代表党以来，每个州的选举结果都是最重要的因素，其席位数量取决于每个州的人口规模。 因此，代表美国各州人口规模的失真的美国地图通常会在美国大选中使用，而不是使用具有地理准确性的地图。 这些地图中的许多地图都使用正方形来表示每个州，其面积代表其选举人的票数，但仍然可以看出每个州之间的地理关系( 数据可视化图表与产品 )。

每个人口群体的意见

《卫报》和《金融时报》的以下两张图表都试图根据人口统计数据来解决英国退欧投票的趋势。 例如，年轻的选民，学位持有者和收入较高的选民压倒性地选择留下，但年轻的选民在总投票数中所占的比例很小，对结果的影响较小。

随时间跟踪意见

人口统计群体对支持方的变化

《纽约时报》从2004年开始根据家庭收入水平反映了支持政党的转变。 此外，尽管低收入群体在历史上一直支持民主党，但教育水平与2016年大选支持党的转变甚至更多相关。 泰晤士报通过各种人口统计属性(例如种族，性别和教育)跟踪了这些变化。

人口特征和结果变化

《经济学人》团队使用YouGov的数据来说明自2019年以来，各方的支持者如何根据2019年5月提出的交易重新定位自己的观点。

摇摆县的情节及其程度

《华尔街日报》根据每个县对政党的支持速度以及自2012年以来的变化情况来绘制每个县。下图说明了2016年总统大选如何导致许多摇摆县。

历史上每个州的支持方

摇摆不定的州在对任何一方的积极支持者之间没有很大差距，对预测美国总统大选具有重要意义。 《华尔街日报》(Wall Street Journal)下图显示了各州支持者之间各州之间的历史差异。 该图可以告诉我们支持对每个州都重要的政党和政治方向的趋势。

随时间映射各种民意调查结果

选民的意见不仅会在每次选举之间发生变化，而且可能在单个竞选期间发生巨大变化。 与《电讯报》从顶部开始的民意测验追踪器类似，《卫报》绘制了2015年英国大选的各种民意测验结果，并从这些情节中绘制了中位数线，以估计整个竞选期间各方的立场。

从2016年开始，每个县在2018年如何变化

《卫报》使用不断变化的箭头地图，通过显示自上次大选以来各地区在2018年中期选举中的变动情况，来说明民主党在众议院的利益。 民主党人在2016年唐纳德·特朗普(Donald Trump)统治的广大地区增加了支持。较大的蓝色箭头在2016年表示更大的遗憾，而通过民意调查追踪这些速度可能暗示下一次选举结果。

产生误导性投票结果的机会

盖洛普(Gallup)在1948年预测，有49.5％的公众将投票选举托马斯·杜威(Thomas Dewey)为总统，但真正的结果几乎是相反的：杜鲁门(49.6％)和杜威(45.1％)。 根据民意调查数据，《芝加哥每日论坛报》发表了著名的标题“杜威击败杜鲁门”。 尽管盖洛普(Gallup)提到他们的准确性在60年代后大为提高，但在包括2000年，2012年和2016年在内的最近选举中，他们的说法是错误的。

如今，选民已经上网，可以更紧密地访问各种媒体所进行的民意调查等信息。 如果民意调查结果表明您的支持方将获胜，这些民意调查结果可能会阻止您去投票。 对候选人而言，更紧密地访问数据也至关重要，这是至关重要的-最后一刻的故事会极大地影响选民的思想，并且随着现代竞选活动越来越在线化，对实时数据的访问越来越多，这一速度越来越大。

说明摆动参与者意见的可能性

经济学家团队使用“三元”图代替通用的二维图来获得英国脱欧民意测验数据–他们的尝试是根据受访者的React描绘民意测验参与者在未达成协议的情况下如何支持留任 ， 休假或休假的立场以及他们根据人口统计资料改变看法的可能性。

选举也并不总是能准确地代表民意

《经济学人》的分析团队根据各种民意测验运行了一个模型，以表明如果所有美国人都投票，2016年美国大选将会如何。 模拟表明克林顿将赢得特朗普，这是根据在实际选举之前的民意测验得出的预测。