The Scenario So Far

Imagine you're a data scientist tasked with estimating the average height of pine trees in a specific forest, as illustrated below. While you manage the planning and analysis from your workspace, an enthusiastic hiker is responsible for gathering the data on tree measurements.

Instead of simply directing the hiker to "randomly choose 20 trees," it's crucial to provide clear and effective instructions.

How tall are the trees in this forest? Photo by Dan Otis on Unsplash

Providing Clear Directions

When I ask my students to enhance their instructions for the data collector (the hiker), they often suggest the classic method:

"Assign a unique number to each tree and then randomly sample those IDs."

While this is an improvement, it still falls short of a professional standard. Experienced statisticians understand that the finer points matter significantly.

The Devil is in the Details

For a statistician, that seemingly straightforward directive quickly unfolds into a myriad of considerations:

"Dear hiker, I apologize, but you will need to traverse the entire forest to assign unique IDs to each tree… and I'm relieved to know you enjoy hiking!"

Before we finish that thought, a flurry of questions arises:

• What type of paint should we use? (Note this!)
• How many buckets of paint will we require? (Document this!)
• What if we run out of paint mid-project? (Make a note!)
• Where's the closest place to restock on paint? (Write it down!)
• How will we store the paint efficiently during our work? (Note this!)
• How do we ensure we can find the trees again if their numbers are drawn? (Write it down!)
• Do we need assistance or equipment for this task? (Document this!)
• What if we miss a tree? (Make a note!)
• How long should we anticipate for this task? (Record this!)
• What if the hiker decides to leave before completion? (Note this!)
• What if rain washes off the paint? (Make a note!)
• What will be our overall budget? (Document this!)
• How will we adjust if costs exceed our expectations? (Write it down!)

Alternative Sampling Methods

As you refine your plan, you may realize your ideal method is impractical. In such cases, you have a couple of choices:

1. Select a different sampling method. SRS isn't the only technique, and it may not be the best fit for your objectives. While SRS is beginner-friendly, if it proves too costly, consider advanced methods.
2. Ignore the challenges. This common shortcut involves proceeding as if certain assumptions are valid, despite knowing they aren't. This is a precarious choice and only the decision-maker can authorize it.

Now, if your hiker is enthusiastic enough to offer a fixed rate for the project, you can afford to have them paint all the trees. Next, you can plan to randomly select 20 trees.

How can we "just" pick 20 trees? One straightforward approach is to wait until all trees are labeled, then input their IDs into a spreadsheet, apply the =RAND() function in the adjacent column, sort by random number, and choose the top 20.

However, this leads to more questions:

• Should the labels be numeric or otherwise? (Document this!)
• What if the labels are difficult to read? (Note this!)
• Is our random number generator truly random? (Write it down!)
• Are we considering all technical needs during the project? (Document this!)

Once we have the 20 selected IDs—where will they be saved? (Record this!)—we must prepare to instruct the hiker to possibly retrace their steps. Ideally, we should have devised a strategy to optimize their time and effort.

How will the hiker conduct the measurements? (Document this!)

Which tools should we use? A hypsometer? (Make a note!)

How will we ensure the hiker is adept in using the equipment? (Write it down!)

What if a tree is on uneven ground? (Document this!)

How will we record the measurements? (Note this!)

It's also crucial to consider what additional information might be valuable to collect, such as time of day, tree characteristics, or proximity to other trees. (Discuss with an expert before proceeding!)

If we are gathering various data points for each tree, how should we organize that data? (Make a note!)

To avoid complications, we must ensure our instructions are thorough. A poorly detailed plan will lead to frustration for both the data collector and the planner.

The Importance of a Sampling Plan

After meticulously outlining the details (write it down!), the end result is known as a sampling plan. This document encapsulates the practical elements of data collection, and neglecting to create one can lead to costly mistakes.

Now that both the theoretical sampling procedure and the practical plan are documented, they can be unified into a comprehensive sampling scheme.

Sampling Plan + Sampling Procedure = Sampling Scheme

When purchasing data, always request the full sampling scheme to ensure clarity on what the data represents. For further insights on managing inherited data, including purchased datasets, refer to my additional resources.

A true professional would never attempt data collection without a complete sampling scheme in place. Unfortunately, the theoretical focus of many data science courses often leaves graduates unprepared for practical challenges. Those of us with experience can advocate for a focus on the practical elements because, without solid details, the complex mathematics mean little.

Thanks for reading! Interested in learning more about AI?

If you enjoyed this discussion and seek an engaging AI course suitable for all levels, check out my course designed for your enjoyment.

P.S. Have you tried tapping the clap button on Medium multiple times to see the result? ❤️

Learn more about the importance of data design

Liked the author? Connect with Cassie Kozyrkov

The first video, "Developing a Sampling Plan," offers insights into crafting effective sampling strategies, highlighting the steps involved in the planning process.

The second video, "Essential Market Research Tips: Documenting Your Sampling Plan," provides essential tips for documenting your sampling strategy effectively.