Tip: Use *experiment to run better studies with balanced groups

Picture this: You just finished running a study with 100 participants, each randomly assigned to one of three interventions, to compare how effective they are. You’re excited to analyze the results, but when you open your CSV file, your heart sinks. Group A has 45 people, Group B has 32, and Group C has only 23.

Your beautiful research design just got a lot more complicated.

If this sounds familiar, you’re not alone. Unbalanced groups are a common headache in research, and they can seriously mess with your statistical analysis. The good news? GuidedTrack’s *experiment feature solves this problem automatically.

What makes *experiment special?

In most survey platforms and research tools, getting balanced groups can be a nightmare. You either have to manually track participant counts, use complex external randomization services, or just cross your fingers and hope for the best.

GuidedTrack’s *experiment feature handles all of this automatically by ensuring that group sizes never differ by more than 1 person. Here’s the basic concept of how it works. Suppose you have two groups in your study. Then, *experiment makes sure that the first two participants end up in different groups from each other (selected at random). The next two people also end up in different groups from each other. And so on. Each pair is randomized to one of the two groups, keeping the size of each group balanced.

Real examples of *experiment in action

🧠 Testing Different Meditation Techniques

Let’s say you want to compare three approaches to stress reduction:

*experiment: Meditation_Study
	*group: Visualization
		For the next 10 seconds, visualize yourself in a warm, happy place.
		*wait: 10.seconds

	*group: Writing
		For the next 10 seconds, write about a warm, happy place.
		*question: My warm happy place is...
	
	*group: Control
		For the next 10 seconds, just sit quietly and breathe normally.
		*wait: 10.seconds

With 90 participants, you’ll end up with exactly 30 people in each group. No more, no less.

📱 A/B Testing App Interfaces

Product teams can benefit from using *experiment to test different user interfaces:

*experiment: Button_Color_Test
	*group: Red_Button
		--[...Set the button color as red...]
	
	*group: Blue_Button
		--[...Set the button color as blue...]
	
	*group: Purple_Button
		--[...Set the button color as purple...]

Whether you get 47 users or 150 users, each version will be seen by roughly the same number of people.

Advantages of using *experiment

Statistical Power: Balanced groups give you the best chance of detecting real differences between your conditions. Unbalanced groups can hide important findings or make you think you found something when you didn’t.

Easier Analysis: Most statistical tests work best with equal group sizes. You won’t need to worry about weighted analyses or explaining why your groups are different sizes.

Professional Results: When you present your findings, balanced groups look intentional and well-planned (because they are).

When to use *experiment

Use *experiment when:

  • You’re comparing different treatments or conditions
  • You plan to analyze differences between groups
  • You want balanced sample sizes
  • You’re running A/B tests

You might not need it when:

  • You just want variety in your content
  • You’re showing random examples or scenarios or you have a very complex randomization scheme (use *randomize instead)
  • Balance isn’t important for your goals

Syntax of *experiment

*experiment: Your_Experiment_Name
	*group: Group_A
		-- Content that only participants assigned to Group A will see

	*group: Group_B
		-- Content that only participants assigned to Group B will see

When you examine the CSV file, you’ll see a column labeled with your experiment’s name in the header row. Each row beneath it indicates the group assigned to that particular run.

Tips for creating studies with *experiment

  1. Name your experiments clearly: “Onboarding_Flow_Test” is much better than “Experiment_1” when you’re looking at data six months later.
  2. Keep groups meaningful: Don’t create 12 tiny groups when 3 larger ones would give you better statistical power.
  3. Document your hypotheses: Before you launch, write down what you expect to find. Your future self will thank you.