Why is it that when we do user research, the sample size is either 50 or 500? Why not 300, or 20 like that? The author made a detailed analysis of this problem through his official account as an example, and explained the theoretical basis behind it, and shared it with you.
Instructors often say that if you don't investigate an issue, you stop speaking on an issue.
It is necessary to pay attention to investigation and oppose nonsense.
Today, a friend is asking me about the relationship between qualitative and quantitative research, and why is qualitative research 50, not 30 and not 100?
Is there any theoretical support?
I was asked at the time, and I was embarrassed to reply hahaha, good question.
After lunch, rethink the question and do a simple combing.
Qualitative Research:
There are several key indicators to be clarified in the qualitative research of purchased users.
1. Who is the user?
2. Where is the user?
3. What stage does the user belong to when using the product?
For example,
I'm writing an official account, and my product is to output something related to the robot track.
So how do I do qualitative research?
1. Who is the user?
Through the background data, you will find that the ratio of male to female fans of my official account is almost 4:1
Then I should follow this logic when I do research, the ratio of men and women.
Who the user is also includes age, occupation, income, and so on, and the relative proportion of that group should also increase.
Then I will focus on the age range of 26-45 in my interviews.
2. Where are the users?
Or take my official account as an example, then the focus of my interview is in these places.
Continuing to dismantle, the Guangdong interview focused on Shenzhen, and the Jiangsu interview focused on Suzhou.
3. What stage does the user belong to when using the product?
1. Unused people
2. People who have been used
3. Heavy experience
Decompose by stage and then by region and user
3 (Stage) x2 (Gender) xN (Region Selection) xS (Age Selection)
Follow my opinion
3x2x4x2 = 48 people
This has not been adjusted according to the proportion of fans, so is 50 reasonable?
The research of new products can also be used as a reference, and it is also applicable.
Quantitative Research:
A survey sample of 500 does not represent valid data recovered, so why 500?
We may have been exposed to some statistics when we were in the reading stage, and there is a probability problem in statistics.
I don't do popular science, and the sample size actually has several key indicators.
1. Confidence intervals
2. Confidence level
3. Population size
4. Standard deviation
Interested in Baidu on your own
There's a concept in statistics called the Z-score.
The Z-score is a concept in statistics that indicates how many standard deviations your score has from the mean. When calculating the sample size, the Z-score represents the confidence level you want.
Common confidence levels are 90%, 95%, and 99%, which correspond to Z-scores of approximately 1.645, 1.96, and 2.576, respectively.
The Z-value table of the positive Tai distribution is ready-made, which can be seen in the figure below
How do you look at the normal distribution Z-value table?
For example, we want to find out what is the Z-score at a 90% confidence level?
A confidence level of 90% means that 90% of the population is covered in the middle, leaving 5% at each end (100% – 90% = 10%, half at each end, so it's 5%).
In a standard normal distribution, this means that you need to find a Z-score with a cumulative probability of 95% (50% + 45%). Because 50% is covered from the leftmost end to the middle, plus 45% from the middle to your Z point.
Use a normally distributed Z-value table
Find the corresponding rows and columns, and then determine the intersection points along the rows and columns, and you can find the corresponding Z-score. If the table is the cumulative probability from the far left to point Z, you'll find a Z-score with a cumulative probability of around 0.9500.
It seems that there is still no talk about the sample size, don't worry.
When you know this information, you will naturally understand a formula
1) Confidence Level (Z-Score): How much confidence do you need? This determines your Z-score.
2) Expected effect size (P): This is usually the smallest change or difference you want to be able to detect.
P = 0.5 is usually used as a conservative estimate, as this gives the maximum sample size.
3) Standard Deviation (e): The range of variation in the expected answer.
If you are unsure, the acceptable margin of error e can be used as a conservative estimate of 0.05 (i.e., 5%).
Let's say we want to calculate the following scenario:
The confidence level is 95%, so Z = 1.96
Projected effect size (proportion) P = 0.5
Acceptable margin of error e = 0.05 (i.e., 5%)
Let's substitute these values into the formula:
N = (1.96² × 0.5 × (1 – 0.5)) / 0.05² = 384.16
According to calculations, the required sample size is approximately 384.16 when the confidence level is 95%, the expected effect size is 0.5, and the acceptable margin of error is 5%.
Calculated with a sample calculator, the user base is hundreds of thousands, and the effective sample size of 382 can be covered.
The user base is 1 million, and the effective sample size is 384.
The user base is 100 million, and you will find that it is still 384.
Amazing no?
Without investigation, how could I be talking nonsense.
more than
This article was published by Everyone is a Product Manager Author【new bo】,WeChat public account:【new bo】,Original/authorized Published in Everyone is a product manager,It is forbidden to reprint without permission.
Image from Unsplash, based on the CC0 license.