Generating hypothesis is the 2nd part of the experimental method series. After a “right” question is identified, the next step is to form hypotheses based on the question.
 |
| The 2nd step of the experimental method, generate hypotheses |
Why hypothesis
A hypothesis is a threshold you defined to judge the result of your research or data analysis. Maybe it is easier to explain this concept with some examples.
Take dating a girl for an example (I like to use dating as an example because most of us have been in that process at least once): you met her a couple of times, and you are wondering if she is interested to become your girlfriend. Here, this is your question. So you picked up the first flower you saw by the road and started to count the petals: she will, she won’t, she will, she won’t… and so on until you got to your last petal and it is either a yes or a no. Regardless of the methodology, you have your hypothesis: “no, she won’t be your girlfriend”. This is your null hypothesis and “yes, she will be your girlfriend” is your alternative hypothesis. The answers yes and no are your decision point, your threshold. And based on this decision point, you can continue with your research, ask the girl if she would like to be your girlfriend.
 |
| Null and Alternative hypotheses - using dating as an example |
Another example here, about making a business decision. How do you know if your business has a market? “Does my product/service have a market?” is a valid question. Normally you would then go on with market research like questionnaires, provide a service trial or product demo and see how many customers are willing to pay for your product. But how many customers who are willing to pay for your product is enough for you to claim that you have a market? 1 out of 100 or 40 out of 100? So your null hypothesis would be there are less than 40% of people who are willing to pay for the product thus your business doesn’t have a market; and the alternative hypothesis is there are more than 40% of people who are willing to pay for your product/service so there is a market for your business.
Generate hypotheses
Before we start talking about how to generate hypotheses in a more systematic way, from the examples above, we know that people naturally have hypotheses in mind as soon as the question is asked. No matter whether that is, if the girl you date will say yes or no to your question, or if your business has or has no market fit.
These comments/answers you have in mind are your hypotheses. Whether If you are aware of the answer you have is potentially a hypothesis is something else.
Before you check if the answers can make a good hypothesis, please write it down. Because this answer would be the base of either your null hypothesis of one of your alternative hypotheses.
Generate hypotheses is the pre-step of experiments and is the baseline when you analyse your data
If you want to utilize this experimental method as a way to solve a problem you have in business or product, it becomes important for you to write down your hypotheses. Because based on the hypotheses we are going to continue the following two steps in the experimental methods: the research/data gathering and the data analysis.
The more precise your hypothesis is, the easier for you to design an experiment/research for it. You have written down the hypothesis, and as I have mentioned your answer could be either null or alternative hypotheses. Now is the time to refine your hypotheses and define either of the null or an alternative hypothesis that your answer is.
Null Hypothesis
In a few words, a null hypothesis is a scenario so that there is no association between your observing object/variable and another object/variable. And a good null hypothesis should be well defined and exact, here’s what Ronald Fisher, “a genius who almost single-handedly created the foundations for modern statistical science”, thinks of a good null hypothesis:
…the null hypothesis must be exact, that is free from vagueness and ambiguity because it must supply the basis of the ‘problem of distribution,’ of which the test of significance is the solution. – Ronald Fisher
As Fisher mentioned the null hypothesis must be exact. You can ignore the part when he mentioned the basis of the “problem of distribution” for the moment, this will be brought up again when we use a numeric example to go through the experimental method. Meanwhile, once the null hypothesis is defined, then an alternative hypothesis(es) will also be written down.
Alternative Hypotheses
An obvious alternative hypothesis is usually to assume that there IS an association between your observing object/variable and another object/variable. Or in another word, the opposite of the null hypothesis.
Some basic rules for the hypotheses, although there should be only one null hypothesis in any of your questions there can be more than one alternative hypotheses. Also, when generating both the null and alternative hypotheses you need to make sure the null hypothesis and alternative hypothesis always should be mutually exclusive. Or if you are familiar with the MECE principle from Barbara Minto at McKinsey & Company, you can view this mutual exclusiveness of two hypotheses a more strict rule for that.
And go back to the dating example I gave at the beginning of this article, the null hypothesis of the girl saying no is exactly a mutually exclusive hypothesis than the alternative hypothesis when the girl saying yes to you.
Above is a quick grasp of what hypotheses are and what they are used for. In the later posts, I will spend more time diving into the idea of what is a null hypothesis and what is alternative hypotheses, how to generate the hypotheses, and how to reject the hypotheses.