As other answers have pointed out the half-life of a substance is based on probably. A particle has a certain probably that it'll decay at any given moment. If you have 200 particles with a 1/100 chance per second to decay, after the first second, on average two will have decayed. If you only have 2 particles with a 1/100 chance per second to decay, it's much less likely that either of them decay. 

To illustrate this I wrote a simulation using p5.js. There are 1000 particles with a 1/1000 chance per frame (1/30 of a second) to decay. Once half the particles have decayed the "Measured HL" is updated. The measured half-life of these particles is about 25 seconds.

Note, once you get to just a few particles the half-life is more likely to deviate from the norm since the sample size is so low.

https://editor.p5js.org/d5c4b3/sketches/WbBfFnynj