When the difference really makes a difference

📍On business resolutions: As a student who has been educated at the university and wrote two books in the field, I would like to share my knowledge on bite-up titles to help you navigate a database and AI confidence and clarity.

🔗Of my friendly professional: Once in each heading, I will share the perfect code, usually packed as the affected assistant works that you can unite easily on your work transmission.

: You are a CEO of Retail Chain with two departments, and B. You have reviewed a quarter report, when the bar chart shows that 80 of 100 scores in 85 stores. If you retrieve 75 scores.

What if I have told you that in one form, this action can cost your company millions, and in the other situation, it is the right travel?

The difference between these two conditions are not in the numbers you see – they are in numbers that do not.

🎯in the next 10 minutes, you will learn:

• How many different business facts can hide behind the same bar chart
• Three active steps to find a full story and avoid the most unfair description

The problem with summaries

Business decisions often depend on simple summaries shown in the bar or charts line:

• Ratings across products
• Customer Satisfaction
• Human participation in all parties

But summaries such as these hide sensitive information – it and the details that can make or have broken your next.

Let's go back to the store example. If you think a chart comparing the store a and keeps B, what do you see? It is probably something like this below: two bars, taller than the other.

Here's the twist: Three different business conditions – each you need a different decision – can produce the same bar chart 🤯.

🔎ready to see what your data doesn't tell you?

The bar chart hides – the rest of the story

Let us look at three different business facts that can hide behind the same bar chart.

Status 1: Little sample, little variations

In Ethony 1, both stores have small sample sizes (n = 50) and Low variance (standard deviation = 5).

This information is not visible on the bar chart. But if we switch to a different graph – box of box dispersation-Ing me per customer's points such as a point, and you can see the calculation test result shown in the corner.

The above graph tells us:

• Customer scores are well integrated next to each stores.
• The skirt with 5 points in the middle of the shops are always visible.
• Mathematical examination (Anatha) confirms the difference is real, not just the opportunity.

💡Key Understanding: In this case, you will be right to re-recycle and maintain a practice and invest in Steat B development.

📦Application Tip for Others: The above graph is easy to make with the code below. In addition to customizing the look, you can choose between different statistics for your data, too. Pls checked MLARENA Docs in GitTub for details.

from mlarena.utils.plot_utils import plot_box_scatter

fig, ax = plot_box_scatter(scenario_a, 
                           x='store', 
                           y= 'satisfaction', 
                           show_stat_test=True, 
                           stat_test='anova',  
                           palette = colors)

Status 2: Small sample, great differences

In the case of 2, both stores still have small sample size (n = 50) and similar scores in the same (80 store a, 75 store b). But now, customer satisfaction scores have Top Differences. This changes the story very much:

• While the bar chart will look so in two cases, from the box above-circles you can say that data points can spread more about the situation 2.
• The difference between two stores is now difficult to distinguish from random noise.
• It is compatible with this method indicated in the system, math analysis indicates the difference It is not mathematical.
• Even if the methods are the same as the nature of 1, we cannot conclude with conviction That store at Apperforms B.

💡Key Understanding: The same difference means a difference you can tell the perfect stories in terms of data irregularity.

What should you do with noisy data?

How do data-operated decisions, where your data is noisy (ie, with maximum variations)? Status 3 provides feedback.

In the case 3, we keep the same top variables as a condition 2 but highly increased the sample size. This reflects the power of great detail:

• Data Points are always widespread (the same variety as a condition 2)
• However, a large sample size provides a lot of mathematical power
• For many data points, we can now separate the signal from noise: math analysis shows a difference is equally important in spite of high variations
• A larger sample gives us confidence that keep apperforms b shop b

💡Key Understanding: When the difference is high, the sampling size can increase our power that gets a real difference.

📦Tip with data expert and: I will introduce the useful use of the use and sensitivity ability in the equal heading of biting.

When an important effect is not a big deal

Match Status 1 including Status 3Can you say that as they both show a difference of 5 statistics statistically, two conditions are actually the same?

The answer is larger No ⛔

• Status 1:
  • 5 point differences invent 100% of normal deviation – a very strong effect.
  • 👉 Suggest a The main difference of work appropriate to be repeated immediately.
• Status 3:
  • Only 5-minute point differences 25% of normal deviation – minimum result.
  • 👉 Showing only a modest benefit That may not bring major changes.

💡 Key Understanding: The value of math tells you that the difference is real. EFFect size Telling that the difference is greater enough to have a business.

📦Tip with data expert and: You guess, I have more useful use of the size of a disciple size to share with you in the future article.

💡Key Understanding: Don't take everything It is equitable The results deserve the same response – size to effectiven the resource allocation.

Put everything together

Acommunicated Key Taking and Modes to Make Business Decisions:

🚫 What you do not:

• Don't make decisions based solely on the difference
• Don't think what it means in the same business means of business

✅ What Should You Do:

1. Always ask for distribution information aside means (eg
2. Ask for a test of mathematical importance Before concluding that a visible difference is working
3. Ask for the size of the read Understanding that mathematically important differences cost the action costs of action

Bonus point: When the consequences are different from high variety, think about collecting the big samples to increase mathematical power and bring clarity.

The row below: 5-Point means that the same difference can allow a quick action (Status 1), ITual Comprehension (Status 2), or to ensure the act of high confidence but humble impacts (Status 3). Obedient Data variation, Mathematical importancebeside EFFect size prevents a negative translation of your business metric metrics.

🔮 What is following: I will write articles with great size showing important ideas on data and AI to make business decisions. EFFect size, math tests including Mathematical force What we touched in this article is everything in the list. Let me know what else you would like to see next 🤗

I write about data, ml, and AI solve problems. You can also find me in 💼linkedIn | 😺github | 🕊️TWITTER /

Unless otherwise noted, all photos are the author.