Selecting procedures

Think of "Selecting procedures" like choosing the right tool from a toolbox. If you want to hammer a nail, you grab a hammer. If you want to tighten a screw, you grab a screwdriver. You wouldn't try to hammer a nail with a screwdriver, would you?

In Statistics, our "tools" are different statistical procedures (fancy ways of analyzing data). When you're given a problem or a question, your job is to figure out which statistical tool is the best fit. This means looking at:

• What kind of data do you have? (Are they numbers, or categories like 'yes/no' or 'red/blue'?) This is like knowing if you have a nail or a screw.
• What's your goal? (Are you comparing two groups? Looking for a relationship? Trying to estimate something?) This is like knowing if you want to hammer, tighten, or measure.

It's all about matching the problem to the perfect solution!

Let's say you want to know if a new brand of fertilizer makes your tomato plants grow taller than your old fertilizer. You plant 10 tomatoes with the new fertilizer and 10 with the old, and after a month, you measure their heights.

1. What's your goal? You want to compare the average height of plants using the new fertilizer to the average height of plants using the old fertilizer.
2. What kind of data do you have? You have actual measurements (heights in inches), and you have two separate groups (new fertilizer vs. old fertilizer).
3. Which tool to use? Since you're comparing the means (averages) of two independent groups, and you have measurement data, you'd likely choose a two-sample t-test for means. This statistical procedure is specifically designed for this kind of comparison. If you tried to use a procedure for categories, it would be like trying to water your plants with a wrench!

When you're faced with a statistics problem, here's how to pick the right procedure:

1. Identify the Question: What is the problem asking you to find out or prove? Are you comparing, estimating, or looking for a relationship?
2. Count the Groups/Variables: How many groups are you looking at? Are you comparing two groups, one group to a known value, or looking at the relationship between two different measurements?
3. Determine Data Type: Is your data quantitative (numbers you can do math with, like height or age) or categorical (groups or labels, like 'yes/no' or 'favorite color')?
4. Check for Conditions: Does the problem give you information about standard deviations, sample sizes, or if the data is paired? This helps narrow down the exact test.
5. Match to a Procedure: Based on steps 1-4, select the specific statistical test or confidence interval that fits all the clues.
6. State Your Choice: Clearly name the procedure you've chosen, like "Two-sample t-interval for the difference in means."

Just like a detective looks for fingerprints, you look for specific words and phrases in a statistics problem to guide your choice:

• "Compare means" or "difference in averages": This usually points to a t-test (for means) or a t-interval (for estimating the difference in means). Think of comparing two different brands of batteries to see which lasts longer.
• "Proportion" or "percentage": These words scream z-test (for proportions) or z-interval (for estimating a proportion). Imagine asking people if they prefer chocolate or vanilla ice cream.
• "Relationship between two quantitative variables": If you have two sets of numbers and want to see if they move together (like hours studied and test scores), you'll think about linear regression or correlation.
• "Goodness of fit" or "association between categorical variables": When you're checking if observed counts match expected counts, or if two categorical things are related (like gender and favorite sport), you're likely using a Chi-square test.
• "Estimate": If the goal is to find a range of plausible values for a population parameter (like the average height of all 12-year-olds), you'll need a confidence interval.

Don't fall into these traps!

1. ❌ Confusing Means and Proportions: Trying to use a t-test when the data is about percentages.

   • Why it happens: Both involve comparing groups, but the type of data is different.
   • ✅ How to avoid: Remember: Means use T-tests (think 'M' and 'T' are close in the alphabet). Proportions use Z-tests (think 'P' and 'Z' are far apart). If you have counts or percentages, it's a proportion. If you have averages of measurements, it's a mean.

2. ❌ Ignoring the Number of Groups: Using a one-sample test when you're comparing two different groups.

   • Why it happens: It's easy to just see "average" and jump to a one-sample test.
   • ✅ How to avoid: Always ask: "How many populations or groups am I drawing samples from?" If it's two distinct groups (e.g., boys vs. girls, old vs. new method), you need a two-sample procedure.

3. ❌ Forgetting Paired Data: Treating paired data (like before-and-after measurements on the same people) as two independent groups.

   • Why it happens: It looks like two sets of numbers, so it's tempting to use a two-sample test.
   • ✅ How to avoid: Look for keywords like "before and after," "matched pairs," or "same subjects." If each data point in one group is naturally linked to a data point in the other, you should calculate the differences and use a one-sample t-test for paired differences.

4. ❌ Not checking conditions: Jumping straight to the test without thinking if it's appropriate.

   • Why it happens: Students want to get to the math part quickly.
   • ✅ How to avoid: Always pause and ask, "Are the conditions met for this test?" (e.g., Is the sample random? Is the sample size large enough? Is the population standard deviation known?). This is like checking if your hammer is actually a hammer before you try to hit a nail with it.