How much information a sample carries
Today let's work out exactly how much one sample is really telling you about the number you are trying to estimate.
A hospital tries a new treatment on patients who either respond to it or do not. Forty patients take it, and 12 of them respond, an observed rate of 0.3. Call the treatment's true response rate p, the number you do not know and are trying to pin down.
Every candidate value of p explains that sample better or worse. Score each candidate by how well it explains the 12 responses actually seen, a score called a likelihood, and plot that score across every value of p from near 0 to near 1. The curve below is exactly that plot, and it peaks right at 0.3, since 0.3 is the value that explains the data best.
The picture mounts a coin instead of a patient, held at the same 30% rate: its own label talks about flips, not responses, but the shape it draws is exactly the trial's shape, since both carry the identical ratio. Look at how fast the curve falls away once you move left or right of the peak. That fall-off has a shape, sharp or shallow, and that shape is a number waiting to be read off it.
A bigger sample sharpens the curve and shrinks the standard error
The same widget is mounted again below, but now its sample-size slider is yours to move. It still runs on a coin held at exactly 30%, the same rate as the hospital's trial, so at 40 flips it is showing the trial's own 12-of-40 shape, just under its coin label rather than a patient one.
Move the slider from 40 up to 160. Watch two things.
The peak does not move. It sits at 0.3 at every sample size, because the rate behind the data never changes, only how much of it you have.
The curve itself does move. At 40, the readout below the picture prices the standard error, how much the estimate would typically wobble if you reran the trial, at 0.0725. Slide up to 160 and that same readout drops to 0.0362, less than half. The curve has visibly narrowed to match: at 160, values even a little away from 0.3 now score far worse than they did at 40.
So a bigger sample does not move the estimate here. The rate was always 0.3. What the bigger sample buys is a sharper curve, and the smaller standard error that comes with it.