I’m stuck on estimating an average from a grouped frequency table. The table lists ranges like 0–200, 200–400, 400–800, 800–1000, and 1000+, with a count for each range. I know the usual method is to use class midpoints times frequencies and then divide by the total, but I’m not sure that still makes sense when the class widths aren’t equal and the last class is open-ended.
Do I still just use midpoints for each class even if some classes are much wider than others, or should the widths change the calculation somehow? How should I handle the 1000+ group when there’s no upper bound to get a midpoint-what’s the standard approximation there? Follow-up: if the table labels are things like “0–4” and “5–9” (which look like there’s a gap if I think of continuous data), do I need to adjust class boundaries before taking midpoints, or is that overthinking it?
I’m looking for a clear, practical rule I can apply so I stop second-guessing myself on these tables.
















3 Responses
Absolutely-estimate the mean with sum(f × midpoint)/sum(f); unequal widths don’t change that, and for 1000+ you typically assume the same width as the previous class to invent a midpoint (or cap it reasonably), while labels like 0–4 and 5–9 get “continuous” boundaries (-0.5 to 4.5, 4.5 to 9.5) so the midpoints still behave nicely. Think of it like balancing a seesaw by placing weights at each interval’s midpoint-here’s a friendly walkthrough: https://www.mathsisfun.com/data/grouped-data.html.
Rule of thumb: still do Σ(f×midpoint)/Σf even with unequal widths; for an open-ended top class I’d usually assume the same width as the previous one to invent an upper bound (e.g., 1000–1200, midpoint 1100), and for 0–4, 5–9 treat them as continuous with boundaries −0.5–4.5 and 4.5–9.5 so the usual midpoints work. I’m pretty sure that’s the standard exam approach, and this quick explainer has examples: https://www.mathsisfun.com/data/grouped-data.html
I’d handle unequal widths by using frequency densities, then estimate the mean as sum(midpoint × density) divided by sum(density), and for the 1000+ class I’d take a midpoint at 1000 plus half the previous width, roughly. For 0–4 and 5–9 I think the gap is real so I wouldn’t adjust boundaries before taking midpoints, though I’m a bit unsure.