“A Survey needs 400 respondents to be reliable.”
The statement is repeated often enough that it can start to sound like a rule of Quantitative Research.
It is not.
For some decisions, 200 respondents may be adequate. For others, even 1,000 may not be enough. And a Sample of 400 provides little protection if the wrong population was recruited or the Survey does not answer the Business Question.
Sample Size should therefore follow what needs to be estimated, the precision required, the population structure and the Analysis Plan, not a familiar number selected in advance.

400 respondents is not a universal standard, it comes from a particular set of assumptions
The familiar “about 400 respondents” figure usually comes from a Sample Size calculation for estimating a Population Proportion in a large population.
Using:
- 95% Confidence Level
- ±5% Margin of Error
- Expected Proportion = 50%
- A large population
Cochran's formula gives approximately 384.16, rounded up to 385 respondents.
Change the assumptions and the answer changes. At 95% confidence with ±3% Margin of Error, the required Sample is approximately 1,068. For a finite population of 1,000, the same 95% / ±5% assumptions produce an adjusted Sample of roughly 278.
The better question is therefore not “Do we need 400 people?” but “What are we estimating, how precise does it need to be, and which subgroups must we analyse?”

Where does the “400 respondents” rule come from?

One widely used formula for estimating the Sample Size required for a population proportion is Cochran's formula:
n₀ = Z² × p × (1 − p) ÷ e²
where:
Z = the Z-value associated with the desired Confidence Level
p = the expected Population Proportion
e = the desired Margin of Error
Using:
Z = 1.96 for 95% Confidence
p = 0.50
e = 0.05, or ±5%
gives:
n₀ = 384.16
Rounding upward produces 385 respondents for an effectively large or unknown population.
When the expected proportion is unknown, p = 0.5 is commonly used because p(1-p) is largest at 0.5, producing the most conservative Sample Size under this formula.
“400” is therefore a convenient rounding of a particular result—not a rule for every Survey.

Why does a smaller Margin of Error require so many more respondents?

Because Margin of Error is squared in the denominator of the formula.
For a large population with p = 0.5:

• 95% confidence, ±5% Margin of Error → about 385
• 95% confidence, ±3% Margin of Error → about 1,068
• 99% confidence, ±5% Margin of Error → about 664

A relatively modest increase in desired precision can therefore require a much larger Sample.

What does a ±5% Margin of Error mean?

Suppose a Survey estimates that 50% of the population prefers Brand A.
Under an appropriate sampling and confidence-interval framework, a ±5 percentage-point Margin of Error creates an interval roughly around 45%–55% at the stated Confidence Level.
But this Margin of Error reflects sampling uncertainty under the assumed design. It does not include every possible Survey error, such as:

• Biased question wording
• Non-response
• Incorrect recruitment
• Low-quality responses
• Coverage Error
• Weighting Error
• Measurement Error

A larger Sample does not automatically repair a poor Survey Design.

Does Population Size matter?

Yes, but not in the way many people expect.
For very large populations, the required Sample Size does not increase linearly with Population Size when precision assumptions remain fixed.
This is why roughly 385 respondents can arise under the same 95% / ±5% assumptions whether the population contains hundreds of thousands or millions of people.
For a smaller known population, however, Finite Population Correction (FPC) can reduce the required Sample.
The adjustment is:
n = n₀ ÷ [1 + (n₀ − 1)/N]
For a population of 1,000 and n₀ = 384.16, the adjusted Sample is approximately 278 respondents.

Why might 400 respondents still be too few?

Because Total Sample Size is only one part of the design.

1. You need to analyse several subgroups

Suppose the total Sample is 400, but the study needs comparisons across:
Bangkok vs. Upcountry
Male vs. Female
Age 18–24 / 25–34 / 35–44 / 45+
Once the Sample is divided, the number of respondents in each analytical cell becomes much smaller.
If each subgroup needs reasonably precise estimates, Sample Size should be designed at the subgroup level, rather than collecting 400 people in total and dividing them later.
For example:
400 respondents ÷ 4 Age Groups = an average of only 100 respondents per group.
A Base of 100 produces much wider sampling uncertainty than a Base of 400.
The relevant question is:
Do we need precision for the total population, or for each Segment?

2. You need to detect a small difference

If the goal is simply to estimate whether Awareness is closer to 40% or 60%, one Sample Size may be adequate.
If the goal is to distinguish whether Campaign A at 51% differs meaningfully from Campaign B at 54%, the required Sample depends on the Effect Size and Statistical Power.
The 385 formula for the Margin of Error of one proportion does not directly determine the Sample Size for every Hypothesis Test.

3. The Sampling Design is more complex

The basic Cochran calculation assumes a setting similar to Simple Random Sampling.
Cluster Sampling, Multi-stage Sampling and other designs may increase the similarity of respondents within clusters and therefore reduce effective precision.
A Design Effect may need to be applied.
Thus, 385 observations under a Cluster Design do not automatically provide the same precision as 385 observations from a Simple Random Sample.

4. You expect non-response or unusable data

If the analysis requires 385 valid completes, the business may need to recruit more than 385 people because some respondents may fail to complete the Survey or fail Quality Checks.
Always distinguish:
people invited / recruited
from
final usable responses

When can fewer than 400 respondents be enough?

There are many situations.

The population is finite

If there are only 800 or 1,000 eligible customers, FPC may reduce the required Sample.

The decision can tolerate a wider Margin of Error

An Exploratory Survey used for directional decisions may sometimes tolerate ±7% or ±10% precision.
The limitation should be stated clearly.

The objective is not to estimate a population percentage

Some Surveys are designed for:

• Questionnaire Piloting
• Early pattern exploration
• Internal employee pulses
• Hypothesis generation
• Scale testing

Sample Size should then follow the particular Analysis Method rather than the 385 rule.

What is the difference between 100, 400 and 1,000 respondents?

Under simplified assumptions close to Simple Random Sampling and a population proportion around 50%:
A Sample of about 100 produces a Margin of Error roughly around ±10 percentage points at 95% confidence.
A Sample of about 385 produces approximately ±5 percentage points.
A Sample of about 1,068 produces approximately ±3 percentage points.
The key lesson is that reducing the Margin of Error from ±5 to ±3 does not require only a modest increase in Sample Size.
It requires increasing the Sample from roughly 385 to more than 1,000.

A large Sample cannot compensate for the wrong Sample

Imagine a Survey with 5,000 respondents.
That sounds far more robust than a Survey of 400.
But if all 5,000 respondents are followers of one Brand's Social Media account and the findings are presented as “what Thai consumers think,” the main problem is Sampling Frame and Selection Bias, not Sample Size.
A smaller but appropriately recruited Sample may be more useful for the decision.
Always distinguish:
Sample Size - How many respondents?
from
Sample Quality - Who are they, and how were they selected?
BEE Research Knowledge makes the same distinction: Quantitative Research requires a clear target, an appropriate Sample, valid questions and correct analytical bases.

Does 400 respondents mean the Survey is “95% reliable”?

No.
A 95% Confidence Level does not mean:
“The Survey is 95% correct.”
And 400 respondents do not automatically make a Survey “95% reliable.”
Confidence Level describes the repeated-sampling behavior of a Confidence Interval under the assumed sampling process. It is not a universal quality score for the Survey.
Survey results can still be biased by:
Question Wording
Coverage Error
Non-response
Selection Bias
Data Processing
Low-quality Respondents
Incorrect Weighting
“95% Confidence” should therefore always be interpreted with the sampling and estimation assumptions that produced it.

Do not calculate Sample Size before deciding what you will analyse

Before opening a Sample Size calculator, answer six questions:

1. Who is the Target Population?
2. What is the Primary Metric being estimated?
3. What level of precision or Margin of Error is required?
4. What Confidence Level is appropriate?
5. Which Subgroups need to be analysed?
6. What Sampling Design and Non-response should be allowed for?

Then select the Sample Size approach that matches the Analysis Plan.
For a proportion estimate, a Cochran-style calculation may be suitable.
For comparisons of means or proportions, Sample Size should consider Effect Size and Power.
For Regression, Conjoint, MaxDiff or Segmentation, additional method-specific requirements may matter.
Sample Size should not begin with:
“We usually use 400 respondents.”

A decision-led way to think about Sample Size

Start with the Decision rather than the number.
For example:
Decision: Should we launch Product A or Product B?
Metric: Purchase Intent / Concept Preference
Analysis: Compare two Concepts across three Customer Segments
Precision Needed: Detect differences large enough to matter to the business
Sample Design: Each Segment needs an adequate analytical Base
The total Sample may exceed 400 even when the overall market is not especially large, because the requirement comes from the comparisons, not just the total-market estimate.
Another example:
Decision: Estimate overall Customer Satisfaction for an internal pulse.
If subgroup comparisons are limited and a wider level of precision is acceptable, fewer than 400 may be sufficient.
Sample Size is therefore part of Decision Design, not a ritual of Survey Research.

Seven questions to ask before answering “How many respondents do we need?”

1. What are we estimating or comparing?
2. What is the Population Size?
3. What Margin of Error is required?
4. What Confidence Level is required?
5. What Population Proportion or variability is expected?
6. Do results need to be read at Total or Subgroup level?
7. Do Sampling Method, Design Effect or Non-response require adjustment?

If these questions are unanswered, saying “400 respondents” is premature.

The takeaway: A Survey does not need 400 respondents—it needs a Sample designed for the question

The familiar figure of approximately 385 respondents has a legitimate statistical origin, but only under a particular set of assumptions:
95% Confidence Level
±5% Margin of Error
p = 0.5
Large Population
A Sampling Design appropriate to the formula
‍
Change the assumptions and the answer changes.
‍
Need ±3% precision → more than 1,000 respondents
Population of 1,000 → roughly 278 may be sufficient under FPC
Need several precise Segment comparisons → the total may need to exceed 400
Most importantly: A larger Sample cannot fix the wrong Sample.
Before asking: “How many respondents does the Survey need?”
ask: “What decision will this Survey support, how precise must the answer be, and for which population or segments?”
Once the Business Question and Analysis Plan are clear, Sample Size becomes a meaningful design decision rather than a borrowed rule of thumb.