Skip to main content

FAQS

General

The British Election Study Team principal investigators are Professor Ed Fieldhouse (Manchester), Professor Jane Green (Nuffield College, Oxford), Professor Geoff Evans (Nuffield College, Oxford), Dr Jon Mellon (Manchester), and Dr Chris Prosser (Royal Holloway, University of London).

We have launched our innovative Data Playground which will allow you to look at BES data and create your own charts, allowing you to look at answers to questions and their relationships without having to have any specialist software.

Historical BES data can be downloaded here.

Users may also find the http://www.besis.org website useful.

Ethnic Minority British Electoral Surveys (EMBES) can be accessed here:

http://www.sociology.ox.ac.uk/research/embes-the-ethnic-minority-british-election-study.html

 

 

The responsible variables all refer to the following grid:

accountability grid

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Respondents are asked to tick each of the parties that are responsible for changes that they identified in the previous questions.

Since respondents can tick as many people as being responsible as they wish, there is a separate variable for each tick box.

So, responsibleNHS_1 equals 1 if a respondent says that the Conservatives are responsible for changes in the NHS and 0  if not. Similarly, responsibleCrime_3 equals 1 if the respondent thinks that the previous Labour government is responsible for changes in the crime rate and 0 if the respondent thinks they are not responsible.

The party contact and priorities questions both work with similar grids that ask respondents to tick all the combinations that apply to them (parties contacting by each method and parties prioritising each issue, respectively).

 

A few reasons:

  • We only have 25 minutes to ask respondents questions and this inevitably means we can’t ask respondents everything we’d like to.
  • We’re focused primarily on respondents’ electoral behaviour. Some questions would be better suited to other surveys.
  • We’re always open to feedback. You can comment on our approach at [email protected]

 

Please contact the following press offices for members of the BES team.

Professor Ed Fieldhouse and Dr Jon Mellon: University of Manchester press office

Professor Jane Green and Professor Geoffrey Evans: University of Oxford press office

Dr Chris Prosser: Royal Holloway press office

Variables

Not applicable and missing values (due to skipping or routing) are coded as missing (. in stata or NA in R).

 

Don’t know values are coded to 9999 in the dataset and labelled as “don’t know”.

 

This means that it is very important to recode “don’t know” values before analysing scales. We chose a deliberately high “don’t know” value so that any diagnostics will immediately pick up on the outlying value.

The responsible variables all refer to the following grid:

accountability grid

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Respondents are asked to tick each of the parties that are responsible for changes that they identified in the previous questions.

Since respondents can tick as many people as being responsible as they wish, there is a separate variable for each tick box.

So, responsibleNHS_1 equals 1 if a respondent says that the Conservatives are responsible for changes in the NHS and 0  if not. Similarly, responsibleCrime_3 equals 1 if the respondent thinks that the previous Labour government is responsible for changes in the crime rate and 0 if the respondent thinks they are not responsible.

The party contact and priorities questions both work with similar grids that ask respondents to tick all the combinations that apply to them (parties contacting by each method and parties prioritising each issue, respectively).

 

Several reasons:

  • YouGov’s existing data on panellists
  • Data from the Electoral Commission
  • Linked data sources

The BES includes much of the information that YouGov routinely collects about all members of their panel:

  • Age
  • Highest qualification
  • Home ownership
  • Marital status
  • Age left education
  • Ethnicity
  • Local authority and education authority
  • Household income
  • Personal income
  • Household size
  • Number of children in household
  • Preferred daily newspaper
  • Vote choice in 2010
  • Religiosity
  • Religious denomination
  • Type of organization worked for
  • Big 5 personality measures

Please check that the problem is in the original data, not in a new coding or manipulation. If the problem persists, please write a detailed explanation of the problem and send to: [email protected]. We will seek to reply as soon as possible and are grateful for any assistance in ensuring the BES dataset is 100% free of errors.

  • Age
  • National Statistics Socio-economic Classification (NS-SEC)
  • Parent’s NS-SEC
  • Highest educational qualification
  • Home ownership
  • Marital status
  • Age left education
  • Ethnicity
  • Local authority and education authority
  • Household income
  • Personal income
  • Household size
  • Number of children in household
  • Preferred daily newspaper
  • Religiosity
  • Religious denomination
  • Type of organization worked for
  • Attending a private primary or secondary school
  • Big 5 personality measures

It is well documented that survey respondents often like to give answers that will make them look better. For instance, surveys almost always show a higher proportion of people who claim to have voted in the last election than actually did. This is a problem if we want to understand why people vote. We might just be predicting who will claim to have voted.

Social desirability measures are questions that almost certainly don’t apply to anyone but do sound like desirable statements.

We asked respondents to tick which (if any) of these statements applied to them (adapted from http://ejop.psychopen.eu/article/view/417/315):

  • I always practice what I preach
  • If I say to people that  I will do something, I always keep my promise no matter what
  • I would never lie to people
  • I always smile at people every time I meet them

The more of these, a respondent ticks, the more likely it is that they are giving answers to make themselves appear good.

 

 

The Moreno scale for Britishness and Scottishness asked respondents whether they felt more British than Scottish.

Our national identity scales instead ask Scottish respondents to place themselves on a 7 point Britishness, Scottishness and Englishness scale. This allows a comparison of more than two identities and preliminary analysis of the Scottish Social Attitudes survey suggested that the two variables separately were better at predicting referendum intention than the Moreno scale.

We also felt that being able to look at the extent to which nation’s identities are felt to be more compatible with Britishness than others.

We coded the most important issue open-ended responses for waves 1-13 manually, albeit supported by machine learning. This process involves a computer assisted coding approach through the codething platform. The software makes recommendations about the most likely categories that responses fall into, but a human coder selects and verifies every entry. In total, 355,000 open-ended responses were coded into categories.

The relevant variables can be found with the stems mii_cat and small_mii_cat. These replace the automated codings with the miilabel, which will be phased out in future data releases. As with previous releases, we strongly recommend that users verify that the open-ended responses match their interpretation of a category, especially when relying on smaller categories in the data.

YouGov routinely collects the postcodes of all its panel members. The British Election Study panel respondents are then coded into various geographies based on these postcodes. Currently, we include variables for three levels of political geography: p_pcon is the respondent’s 2010 Parliamentary Constituency, p_oslaua is the Local Authority and profile_GOR is their Government Office Region. Users can merge in any external data they have recorded at these geographies.

Data for lower levels of geography can be accessed via the UK Data Service. For the 2015 face-to-face survey, LSOA identifiers are available, while for the panel study, MSOA will be available in early 2021.

Profile variables are the demographic variables collected routinely about respondents by YouGov. Having these measures from YouGov’s existing data leaves more space for other important questions on the rest of the survey and avoids asking respondents for information they’ve already provided in the past.

These variables have the prefix ‘p_’. Profile variables are collected at intervals determined by YouGov and therefore should not be assumed to be measured at the time of a wave. Profile information with a wave identifier attached indicates that the profile data was the most up to date available at the time of that survey.

Profile variables include the demographic variables for:

  • Age
  • National Statistics Socio-economic Classification (NS-SEC)
  • Parent’s NS-SEC
  • Highest educational qualification
  • Home ownership
  • Marital status
  • Age left education
  • Ethnicity
  • Local authority and education authority
  • Household income
  • Personal income
  • Household size
  • Number of children in household
  • Preferred daily newspaper
  • Religiosity
  • Religious denomination
  • Type of organization worked for
  • Attending a private primary or secondary school
  • Big 5 personality measures

Yes. It is important to use the relevant weights in order to ensure the sample is representative of the general population, which takes into account the way the survey company generates the sample and the over-sampling of Scottish respondents in the online survey (for example).

Using SPSS: go to ‘Data’ in the menu, ‘Weight Cases’, then select ‘Weight cases by’ and select the weight variable from the left-hand variable list and click ‘ok’. The weighting will be shown as ‘Weight on’ in the bottom right hand corner of the Data Editor screen.

Using STATA: after your stata command include [iweight=W8_2].

More information on the weights provided and their recommended use is available in the documentation accompanying each dataset.

Analysing BES data

We have launched our innovative Data Playground which will allow you to look at BES data and create your own charts, allowing you to look at answers to questions and their relationships without having to have any specialist software.

Not applicable and missing values (due to skipping or routing) are coded as missing (. in stata or NA in R).

 

Don’t know values are coded to 9999 in the dataset and labelled as “don’t know”.

 

This means that it is very important to recode “don’t know” values before analysing scales. We chose a deliberately high “don’t know” value so that any diagnostics will immediately pick up on the outlying value.

Please check that the problem is in the original data, not in a new coding or manipulation. If the problem persists, please write a detailed explanation of the problem and send to: [email protected]. We will seek to reply as soon as possible and are grateful for any assistance in ensuring the BES dataset is 100% free of errors.

Yes. It is important to use the relevant weights in order to ensure the sample is representative of the general population, which takes into account the way the survey company generates the sample and the over-sampling of Scottish respondents in the online survey (for example).

Using SPSS: go to ‘Data’ in the menu, ‘Weight Cases’, then select ‘Weight cases by’ and select the weight variable from the left-hand variable list and click ‘ok’. The weighting will be shown as ‘Weight on’ in the bottom right hand corner of the Data Editor screen.

Using STATA: after your stata command include [iweight=W8_2].

More information on the weights provided and their recommended use is available in the documentation accompanying each dataset.

About the BES data

A panel study is a survey that tracks the same people over time. Respondents are asked some of the same questions over multiple waves in order to see how their views evolve over time. For much more information, see the wikipedia page on panel/cohort studies.

A few reasons:

  • We only have 25 minutes to ask respondents questions and this inevitably means we can’t ask respondents everything we’d like to.
  • We’re focused primarily on respondents’ electoral behaviour. Some questions would be better suited to other surveys.
  • We’re always open to feedback. You can comment on our approach at [email protected]

 

The BES internet panel data is collected by YouGov using an online sample of YouGov panel members.

In addition, after each General Election we also conduct a face-to-face survey using a representative sample of the general population. The 2015 and 2017 face-to-face fieldwork was overseen by GfK, while the 2019 fieldwork was overseen by Ipsos MORI, each time surveying the original 2015 sample.

The 2019 study was designed as a face-to-face in-home survey administered by an interviewer via Computer Assisted Personal Interviewing (CAPI). Fieldwork commenced under the original design but was halted in mid-March 2020, as a result of the coronavirus pandemic. A second phase of research was implemented, following a push-to-web with a postal follow-up approach with the questionnaire self-administered by respondents, either online via Computer Assisted Web Interviewing (CAWI) or on paper (PAPI).

More details are available in the technical report for each survey, included alongside the data on the cross-sectional data page.

Technical

Not applicable and missing values (due to skipping or routing) are coded as missing (. in stata or NA in R).

 

Don’t know values are coded to 9999 in the dataset and labelled as “don’t know”.

 

This means that it is very important to recode “don’t know” values before analysing scales. We chose a deliberately high “don’t know” value so that any diagnostics will immediately pick up on the outlying value.

We name variables in the dataset using descriptive names in camel case.

http://msdn.microsoft.com/en-us/library/x2dbyw72(v=vs.71).aspx

Naming variables descriptively helps to make analysis code more readable than naming variables by their question number. Question numbering is also less practical in a panel study, where a particular variable may change its position across waves.

The Moreno scale for Britishness and Scottishness asked respondents whether they felt more British than Scottish.

Our national identity scales instead ask Scottish respondents to place themselves on a 7 point Britishness, Scottishness and Englishness scale. This allows a comparison of more than two identities and preliminary analysis of the Scottish Social Attitudes survey suggested that the two variables separately were better at predicting referendum intention than the Moreno scale.

We also felt that being able to look at the extent to which nation’s identities are felt to be more compatible with Britishness than others.

We coded the most important issue open-ended responses for waves 1-13 manually, albeit supported by machine learning. This process involves a computer assisted coding approach through the codething platform. The software makes recommendations about the most likely categories that responses fall into, but a human coder selects and verifies every entry. In total, 355,000 open-ended responses were coded into categories.

The relevant variables can be found with the stems mii_cat and small_mii_cat. These replace the automated codings with the miilabel, which will be phased out in future data releases. As with previous releases, we strongly recommend that users verify that the open-ended responses match their interpretation of a category, especially when relying on smaller categories in the data.

Questionnaire Proposals

The innovation module contains items that are only asked to subsets of respondents. This includes questions submitted by BES users plus other experimental questions, and cover topics such as a respondent’s propensity to vote for a given party, whether they discuss politics regularly and if so who with, and how they would feel if a son or daughter married a supporter of a particular party (as a measure of social distance).

Note that these questions may also appear in the main survey in subsequent waves.

The British Election Study Team is pleased to invite members of the BES user community to submit applications to include custom designed content to future waves of the BES Internet Panel Study.

Applications should be no longer than 2 pages (A4, font size 12) not including the questions or instruments which may be appended separately. Applications should set out the justification for inclusion in the BESIP, bearing in mind the criteria set out above. All proposals must include full details of the actual questions or experiments (including wording and response categories), rather than examples of possible instruments.

All proposals must be received at least 4 months in advance of the wave in which they are to first appear. This is to allow adequate time for the board to consider proposals, for the team to work with the proposer and with YouGov to finalise and test the instruments.

A decision will normally be made within two months of submission.

Further details on eligibility, criteria and how to submit your application can be found here.