A week from today ,on Thursday 18 September, 2014, the people of Scotland will have the opportunity to vote on whether Scotland should be an independent country. As it is expected polling companies have measured public opinion in order to assess whether or not Scots will vote in favor or against in the referendum.
When we look at all the polling data publicly available[i] we can see that throughout the past two years most polls have given the No a majority (close to 50%), while the “Yes” option (for independence) has seen an average of support around 37% of interviewees (see figure 1). Of course the tendency over time has changed and the polls have gotten increasingly close in the last month.
Given that what is important is to try to figure out the outcome come September 18th we need to see the evolution of this polling data, and to average of these polls. In order to average the polls we need to consider that each poll is different, and more importantly that each polling house proceeds differently to survey the public and weight the responses. As such I used a statistical technique to average the polls which takes into account these differences[ii].
Figure 2 plots all the available polls (in red the No option and in blue the Yes) without considering non respondents. The blue and red line show the evolution of the averaging of the polls. As it can be seen the gap, which was very wide for a long time has considerably narrowed in the last month. And while the ‘Yes, Scotland’ campaign is still losing the difference is less than 4 percentage points (Yes: 48.35% ; No: 51,65%).
Just to end the analysis, it is worth noting that despite the differences in methodologies between the different polling houses currently polling scots, there doesn’t seem to be any major biases. In fact, according to the estimation no house has a bias significantly different than zero (figure 3); which is not necessarily the case all the time and means that there is overall agreement between all polling houses and where the trend line is.
[i] Sources of survey data: