MarketPredict Investigates: Does Wikipedia Traffic Predict Polls?
Here at MarketPredict we spend a lot of time determining what data is useful for predicting election results. One thing we see often is the value of data varying from election to election, so we make a point of validating data usefulness often. One such data-set we’ve used for some time is Wikipedia candidate page traffic. Will Wikipedia page traffic still be useful in the 2020 political cycle?
Right now, the big election everyone is interested in is the 2020 Democratic Presidential Endorsement contest. Looking at national poll position and Wikipedia traffic we see a bit of a mess:
Major events – debates and major candidates entering the race – are definitely driving Wikipedia traffic. Yet no relationship between poll movements and Wikipedia traffic is obvious. Perhaps the relationship isn’t as useful this cycle?
But what if we evaluate analytically? Looking at correlations between historic polling and current Wikipedia traffic along with historic Wikipedia traffic and current polling:
All correlations are positive, so we know Wikipedia candidate page traffic and polling are related. We also see a spike in correlation with one to four day old Wikipedia traffic, so Wikipedia history has predictive power for polls in the short-run! Thus Wikipedia traffic remains useful, but most of the value resides in the final days of a campaign.
We account for this relationship with MarketPredict, weighting Wikipedia traffic more heavily when it matters. Along with the other roughly twenty data-sets we track, the MarketPredict Team is able to build highly adaptable and reliable campaign simulation models. If you or your clients are looking to better utilize polling information, optimize your media spend or just want to have a better shot at winning your campaign, please let us know!