It's not really arbitrary; it's basic statistics if you want to go out and test your hypothesis and you want to persuade your peers your raw data is as pure as possible.
A core aspect of statistics is to allow people to calculate how accurate their sample reflects the actual population, and conversely, how many samples must one take to obtain a given accuracy. Almost all numbers obtained in real life are samples of the total population. How many females are there it the world? No one can count all 7 billion people, so census takers try to get representative samples from different parts of the world to count, and then use statistics to determine how accurately those samples predict the total population. Same idea if one was trying to figure out a number for the height of the average Londoner, and the typical range to be expected.
Same idea even in reverse: statistics often permit people to decide, given numerical fluctuations in almost all things, how much effort is "good enough"
for one's purpose. There is usually no point spending extra effort and money to do more than "good enough." If I only want to know the percentage of males to females in the world +/- 0.1%, then I need to use a sample size of x; if I want to know +/- 0/01% then I need to use a larger sample size of y. If my goal doesn't care +/- 0.0001%, then there is no point in me sampling the still more people required to obtain this number. The only "idea" would be to count every one of the 7 billion, but why bother unless your question required it.
We wash our hands until they are clean enough for a given purpose, such as eating food vs. surgery, which require different levels of washing,. But there is no absolute ideal. If one continues to wash until one achieves as absolutely clean hands as possible, one would spend hours in the wash room and wash all the skin off.
