TL;DR: In a stats class, a sample percent is count in a cell ÷ sample size × 100 (relative frequency). It describes the sample. A population percent is a different object and usually unknown. Write “of 200 surveyed” if that is what you have. Do not write “of Americans” because the sample felt patriotic.
Relative frequency is part over whole
34 of 200 students walk: 17%. The whole is 200, the number asked, not 34, not the school enrollment of 4,000 unless you sampled all 4,000. If 12 people skipped the question, some teachers use 188 as the whole for that item. Missingness policy belongs in a footnote. Silent 200 vs 188 is a 1-point fight that looks like rounding.
Two-way tables: “what percent of walkers are juniors” uses walkers as the whole, not the class. Conditioning changes the denominator. The same 10 juniors who walk can be 10/34 of walkers or 10/200 of the sample. Both true. The phrase after “of” picks.
Percents versus probability language
17% of the sample walked. Saying P(walk) = 0.17 is a model if you treat the sample as the probability space. Teachers who want a distinction will mark “17% of students” as overclaim if the sample is biased. The calculation 34/200 is still 0.17. The noun on the 17% is the fight, not the division.
Complement: 83% did not walk, if everyone is classified and there is no third category. If “other” exists, 100 − 17 is not “bikers.” Remaining percent is a lump. Split it only with counts.
Margin of error is not a second percent of the same kind
“17% ± 4%” often means ±4 percentage points, so a 13% to 21% interval, not 17% × 1.04. Treating ±4% as relative 4% of 17 (about 0.68 points) is a different statistical product. Intro homework usually wants points. Read “percentage points” if printed. If not, ask. Do not multiply 0.17 × 0.04 and add that as a “margin.”
A percent calculator will not produce a margin of error from 34 and 200. It will produce 17. You still need the SE formula if that is the unit. Keep tools in their lanes.
Percents that should not be averaged
School A 30% yes on n=10, school B 20% yes on n=1000. The unweighted average of 30 and 20 is 25%. The combined percent is (3+200)/1010 ≈ 20.1%. Averaging percents from unequal samples is a celebrity error. Weight by counts. Percent tools do not know your sample sizes unless you go back to counts.
If you only were given percents and not n, you cannot recover the combined percent. Insufficient data. Inventing n=100 for each is a simulation, not a result.
Graphs: pie charts demand 100%
If categories overlap (walk or bus or both), a pie is the wrong picture and percents can sum past 100. That is allowed for “percent who use each mode” as separate yes/no items. It is not allowed for a single-choice pie. Choose the graph to match the whole. Do not force a pie because percents feel round.
3D pies and exploding slices do not change 17. They change how 17 looks. Stats class grades the 17 and the denominator sentence, not the explode.
From percent back to expected counts
If a model says 17% of 200, expected count is 0.17 × 200 = 34. Chi-square later will care that 34 is a count, not a percent. Convert when the procedure wants counts. Stay in percents when the caption wants percents. Switching mid-formula is how 17 gets treated as 17 people.
Write “34/200 = 17%” on one line in the lab. That line is the whole contribution of this article to your report. The rest of the report is about sampling, not about the division.
Percents in abstracts and posters
Posters love 17% without 34/200. A reader cannot judge 17% of n=8 versus n=8000. Always pair the percent with n, or with the count. If space is tight, 34/200 (17%) is enough. 17% alone is a costume. Journals that demand 95% CIs still want the count in the table. The division did not get you off the hook for n.
Stacked bar charts that add to 100% imply a single-choice whole. If your survey was multi-select, do not stack to 100. Use a bar per option with a whole of respondents, each bar allowed to ignore the others. The percent formula stays part/whole; the picture must not invent a partition you did not measure.
Weighted class percents versus survey percents
A teacher’s grade weights are not sampling weights. Do not feed quiz 20% / exam 50% into a relative-frequency pie about how students got to school. Different machines. Survey weights (if a class used them) belong in a later course; intro relative frequency is still count/n. If someone “weights” a survey by shouting, they invented a whole. Invented wholes make invented percents. Stay with counted n until the instructor issues weights in a table.
Attendance when the roster changes on Wednesday
Monday’s class had 30 enrolled and 27 present, which is 90%. On Wednesday two students transferred out and one transferred in, so the roll is 29, and 26 are present, about 89.7%. Averaging 90 and 89.7 as if both percents shared a denominator pretends the school had a stable whole. If the question is “what percent of this week’s seat-days were filled,” add the counts: 27 plus 26 present, over 30 plus 29 enrolled. That is 53/59, about 89.8%. The average of the two daily percents is a close accident here and a lie when one day has a tiny class and the other has an assembly.
Write the counts in the margin before the percent sign. A percent without a count cannot be reweighted later, and the substitute teacher will not remember which Wednesday the roster shrank.
A parent turning a batting average into a percent
The app shows .425 and a parent announces “she hits 42.5% of the time, so she fails 57.5%.” The .425 is hits over official at-bats, a sample relative frequency, not a claim about her career and not a probability the next pitch is a hit. Walks may be excluded from the denominator. A second parent who uses plate appearances as the whole will publish a different percent from the same weekend. Both can be honest if they name the whole. Neither should be averaged with a teammate’s .200 on 5 at-bats to crown a “team average percent” without weighting by attempts.
