Too often, we assume that a failed experiment is a wasted effort. But not all anomalies are useless. Here’s how to make the most of them

Source

1. Check Your Assumptions

Ask yourself why this result feels like a failure. What theory does it contradict? Maybe the hypothesis failed, not the experiment.

2. Seek Out the Ignorant

Talk to people who are unfamiliar with your experiment. Explaining your work in simple terms may help you see it in a new light.

3. Encourage Diversity

If everyone working on a problem speaks the same language, then everyone has the same set of assumptions.

4. Beware of Failure-Blindness

It’s normal to filter out information that contradicts our preconceptions. The only way to avoid that bias is to be aware of it.

anyone who've ever involved in REAL research would know this. first we need a hypothesis, and later through the experiment, prove the hypothesis right or wrong. both right and wrong are a result of the research. research is is only bad if it cannot return any result.

what if the research data all come out wrong?

do you discard the data or do you change the hypothesis, or both?

or do you start a new branch of research on why the data come out so weird?

the 3rd option may prove to be very useful. sometimes it might result in unexpected new discoveries

check your equipment. check your method.

You learn from people's failures. Life's too short to learn from your own mistakes all the time.

if the method is right but the results are wrong, then it IS STILL a valid result.

for example, u hypothesize in the beginning that "eating an apple a day will keep the dentist away". the null hypothesis would be "eating an apple a day will NOT keep the dentist away".

so after studying and researching all ur data, u found that "eating an apple a day will not keep the dentist away". all data is still correct, but u still need to go to the dentist every 6 month. so the result is clear. apples dont do anything. it's the whole mentality of people nowadays that keeps going back to the dentist.

if u're doing a PhD, a null hypothesis is STILL a result.

hypothesis is just a early assumption on a conclusion on ur experiment... it may be right may oxo be wrong...

no matter what the outcome from ur experiment or practical, even disagree with hypothesis, it still a result... once cant exclude that from ur conclusion...

but is fine if u can repeat with other method or technique to get a conclusion which agree with ur hypothesis...

but we have to explain... as long as there be a reasonable explanation that supported with data from experiment or practical, that is a good result...

Yes, what is a failure. You got your results from the experiment did you not, albiet not the one you wanted/expected.

There is no failures in experiments, in the end u will get results which later guide u on how to reset the parameters of the experiment.

Failure to follow the Hypothesis, Experimentation, Conclusion will lead to nothing but chaos of the subject under study.

Hi Fellows Blofeld, jonoave, mycolumn,

How would you reduce the likelihood of control failures with innovations in the Design of Experiments?

Would you recommend Taguchi methods?

Experimental research is not my forte.

Oh I see... So, am I safe to confirm Survey Research in Life Sciences doesn't require experimental design?

This post has been edited by Critical_Fallacy: Nov 27 2013, 04:41 PM

I'm curious on that too.

For example, Abraham Maslow published his paper "A Theory of Human Motivation" in 1943, as we know it today called Maslow's hierarchy of needs. How do researchers validate the existence of universal human needs?

Not sure what is Taguchi method.

First of all, is to check whether the result is consistent with my expectation. Sometimes this step it's good to get feedback from others. For example recently I just finished a script to run some data analysis. The results looked ok to me, but some of my labmates and boss thought some of the numbers were a bit low.

I took a look back at the script and found little bug that takes and process the numbers, where the numbers were being incorrectly reported to the results. So for me, the first step is to take a thorough look at your methods to see if there were any possible errors - that include your reagents.

If the methods and materials seemed fine, then try to analyse why is the results as such. Sometimes unexpected results can be interesting too. This happens very often in phylogenetics studies (study of relationships using genetic data) - e.g. crocodiles are found to be closely related to birds, the divergence of a species etc.

And the checking of method here is important too. There are so much variance and noise in phylogenetic dataset (missing data, fast-evolving genes, horizontal gene transfers etc) that can contribute noise and in turn, produce misleading assumptions.

Thank you for your explanations. When designing physics experiments, we are encouraged to use Response Surface Method to find the optimal process control settings to achieve peak performance. An example of a response surface with two independent parameters (driving factors / inputs) is shown below: