Scientific Method

The scientific method is a form of problem solving that focuses on causes and effects within a specific situation. You can use the scientific method to troubleshoot issues and, by process of elimination, discover the true causes.

Steps of the Scientific Method

1. Define the problem. Make sure that you fully understand the problem, all of the components, the people involved, and what the desired resolution is. You can use these analysis tools to fully define the problem.

2. List variables. Think of all of the possible causes of the problem. Write them down. Use root-cause analysis to discover the various factors involved.

3. Create a hypothesis. Predict the outcome when you test the situation.

4. Control all variables but one. Create a situation in which all of the possible causes are ruled out except the one you are testing.

5. Design an experiment. Identify the problem, your hypothesis, your method, and how you will control variables.

6. Conduct the experiment and record data. Does changing the one factor change the outcome? If not, what other factor or set of factors is most likely to be the true cause?

7. Conclude whether the experiment supported the hypothesis. If so, you have your solution. If not, you can create a new hypothesis, selecting a new variable to test.

Define the Problem

Usually, problems identify themselves: Something fails. It might be a product, service, plan, machine, or even relationship.

Customer complaints about our website double every Wednesday.

List Variables

After you’ve identified the problem, you should list the possible causes for the problem.

• Does our website have double the traffic on Wednesdays?

• Are website problems caused by updates on Tuesday nights?

• Is there a specific type of complaint that shows up mostly on Wednesdays?

• Are customers using a specific web service most often on Wednesdays?

• Is there a bandwidth bottleneck or other ISP problem on Wednesdays?

Create a Hypothesis

After you have identified the possible causes of the problem, you need to choose one possible cause that you want to test for and create a hypothesis about it. Start by identifying the cause that you want to test for.

Are customers using a specific web service most often on Wednesdays?

Then you create a prediction that connects the possible cause to the problem. This is your hypothesis. State your hypothesis as a fact (to be tested).

Customers most often make ticketing purchases on Wednesdays, preparing for events over the weekend, so the problem lies in our ticketing service.

Control Variables

Start by identifying the variables that you want to control:

1. Does our website have double the traffic on Wednesdays?

2. Are website problems caused by updates on Tuesday nights?

3. Is there a specific type of complaint that shows up mostly on Wednesdays?

4. Is there a bandwidth bottleneck or other ISP problem on Wednesdays?

Then figure out how to rule out or control all of these variables:

1. I need to check website analytics to make sure that Wednesday does not have inordinate levels of overall traffic.

2. I need to check with IT to find out what, if any, routine updates roll out on Tuesday nights.

3. I need to review the complaint logs for the last two months to find out if a specific type of issue shows up most on Wednesdays.

4. I need to check with our internet service provider to make sure that our bandwidth is consistent throughout the week.

Design an Experiment

List your problem and hypothesis. Then outline the method you will use to test the hypothesis:

Conduct the Experiment and Record Data

Make measurements according to your experimental design and write them down. If you have many measurements, record them in a spreadsheet so that you can easily search them for patterns. If possible, output different columns and rows of your spreadsheet as graphics (bar graphs, line graphs, pie graphs, and so on) to help you recognize patterns.

Draw Conclusions

First, check to see whether your prediction matched the outcome of the experiment. If it did, your hypothesis has been supported. If not, consider whether the hypothesis should be rejected. Whatever the outcome, think about whether any flaws in experiment design skewed the results.