• Question: how can you prove that your results are acurate?

    Asked by mimi98 to Damien, Rachael, Simon, Suzi, Tim on 16 Jun 2011. This question was also asked by flowerpower11.
    • Photo: Suzi Gage

      Suzi Gage answered on 13 Jun 2011:

      @mimi98
      Thanks for your question, it’s a really good one.
      Often one study by itself may find something interesting, but if someone else cannot repeat this finding, it is probably only by chance, and not an interesting result. So we do what we call ‘replication’ and if a number of different people using different data can find the same result, then this suggests the result is accurate.

      Hope this answers your question 🙂

    • Photo: Rachael Ward

      Rachael Ward answered on 13 Jun 2011:

      Hello,

      This is a very important aspect of science and so a very important thing to ask!
      I do the same as Suzi – firstly I repeat my experiment many times to make sure I get the same answer. If I am concerned about it, I may ask someone to replicate if for me too.
      Depending what you are investigating, you can also ask the same question in a lot of different ways and if all the results point to the same answer this gives you more confidence that your results are accurate.

    • Photo: Damien Hall

      Damien Hall answered on 16 Jun 2011:

      In my part of science, we use something called probabilities to say whether a result is likely to be right or not. (When you ask a question, we don’t know anything about you, like how old you are, but, if you’re at ‘A’-Level age, you might come across them in Maths or Statistics ‘A’-Level.) So, here’s an example. You might know (especially if you’re from the North) that people in the North of England pronounce “put” (like “put it down”) and “putt” (like in golf) the same, but people from the South pronounce them differently. In phonetics I can do measurements of those vowels and find out the scientific properties of the sound-waves that make them up, so that I can attach a number to each vowel. Say that for some person I make them say “put” 50 times and then “putt” 50 times. Now I’ve got 100 numbers standing for 100 vowels, and I want to prove whether that person says “put” differently from “putt” or not. I can take the 50 numbers for “put” and do a statistical test to compare them with the 50 numbers for “putt”. The test will give me a number (the probability) which shows how likely it is that the “putt” numbers are different from the “put” numbers on average.

      But you see I’m only talking about likelihood! In linguistics we usually say that a p-number that shows that 1 “put” number in every 20 is the same as a “putt” number is good enough. That means that we say we have proved that two sets are different even if 1 in every 20 of their members are the same! Actually, it’s not possible to completely prove that something is true or not – it’s only possible to say that it is very likely or very unlikely to be true.
      In other parts of science, the 1-in-20 kind of proof that we take in linguistics is not good enough. Imagine if a 1-in-20 proof was used in medicine – that might mean that, when you tested a new drug, 19 guinea-pigs were cured after taking it but one got worse! So, in medicine, they have a much stricter standard of proof – 1-in-10,000, or even more.

Comments