The Phillips Academy Poll values transparency in the polling process. Below is our methodology created in consultation with faculty from Phillips Academy’s statistics department.
The purpose of all Phillips Academy Poll surveys is to predict the outcome of a particular state-wide senatorial or gubernatorial election in the 2022 midterms. Polls may also compare hypothetical head-to-head matchups of the leading candidates of each party to see how they would fare in November. When weighted by demographic data, these results should paint an accurate picture of the political landscape of the state in question.
Our polls are conducted over a variety of modes.
The sample of landlines used in our polls is selected using Random Digit Dialing (RDD). We randomly select an equal number of phone numbers from every landline block (first 7 digits—area code, local area code, and block number) servicing the state being polled. As a result, all phone numbers designated to the state being polled have an equal chance of being selected.
The sample of mobile numbers used in our polls is purchased from a consumer phone number database.
We typically conduct our polling using a mix of live calls, interactive voice response (IVR) phone calls, and text messages. Candidates are read out loud in the order they appear on the ballot.
The Phillips Academy Poll understands that political preferences can be potentially sensitive information. For this reason, individual responses are completely confidential–in fact, collected data is not associated with particular phone numbers. Furthermore, only the aggregate of all respondents is ever published, thus no individual interviewee or set of political preferences will ever be personally identifiable.
The Phillips Academy Poll’s survey questions are framed as neutrally as possible to minimize bias. When asking interviewees to choose between the candidates, the candidates and their party affiliations are read aloud in the order they would appear on the ballot.
The Phillips Academy Poll’s headline results are achieved by a weighting technique known as raking, or iterative proportional fitting, which makes the results more representative of the state’s population. A computer program is used to calibrate the weights assigned to each demographic group so that the headline results are more indicative of true public opinion. Because the demographics of our respondents will likely not closely align with those of the state, this step corrects for any overrepresented or underrepresented groups in our sample.