Distributions, means, and proportions in survey data. Students use the Profile Explorer and Trends tool to visualize ANES variables and compare groups across election years.
Before you can test a hypothesis about what predicts political opinions, you need to know what those opinions actually look like in the data. A distribution shows how survey responses spread across the possible answer choices. Is most of the public clustered at one end of a scale, or spread evenly? Does the typical respondent land near the middle, or at an extreme? These are factual questions about the data that descriptive statistics can answer — and that any analysis must get right before moving to inference.
A question with a small number of named choices, often with no natural order or a simple yes/no split.
Example: Did you vote in the presidential election? (1=Yes, 2=No, 3=Do not know)
Summarize with: frequencies and percentages for each category. The mean is not meaningful when categories have no numeric order.
Ordinal: choices are ordered low to high but intervals are unequal. Example: 7-point ideology scale (1=Extremely liberal to 7=Extremely conservative). Summarize with median.
Thermometer: runs 0 to 100. Respondents rate a group or candidate: 0=very cold, 50=neutral, 100=very warm. Most continuous of all ANES variable types -- use mean.
ANES samples are not perfectly proportional to the U.S. population — some groups are overrepresented or underrepresented in any given year. Survey weights correct for this: each respondent is assigned a weight that makes their contribution to aggregate statistics reflect what fraction of the real population they represent. When Pulse 2.0 shows you a weighted mean or a weighted percentage, it means the calculation applied those corrections — the number describes what Americans collectively believe, not just what showed up in the specific ANES sample that year.
The percentage missing for a variable is the share of all respondents who did not give a usable answer. High missing rates matter because:
1. If the question was only asked in some years, the missing cases are simply respondents from years when the question was not on the survey — not evidence of confusion or sensitivity.
2. If missing responses are not random (e.g., lower-education respondents are more likely to say "don't know" on policy items), then statistics computed on valid cases may not represent the full population.
Pulse 2.0 always displays N_total and N_valid so you can check the missing rate before interpreting any statistic.
Try it now: The interactive Layer 2 tool lets you pick any ANES variable, make a prediction about its distribution, then reveal the real weighted data. Open it at Layer 2: Distribution Explorer →