Metode Sampling

Aspect	Details
Subject Area	Biostatistics, Epidemiology, Preventive Medicine
Core Concept	How subjects are selected from a target population for inclusion in a study
Why It Matters for Exams	Questions test whether you can identify the sampling method described in a vignette, recognize the bias it introduces, and judge the generalizability (external validity) of the study's findings
Two Main Categories	Probability sampling (every member has a known, nonzero chance of selection) vs. Non-probability sampling (selection is not random)
Most Tested Angle	Identifying which type of sampling was used, recognizing sampling bias, and understanding how the method affects external validity
Keywords	"Randomly selected," "every nth patient," "divided into subgroups then randomly selected," "entire clusters chosen," "volunteers," "referred by other participants," "convenience," "generalizability"

01The Conceptual Foundation

In clinical research, it is almost never feasible to study an entire population. Instead, investigators select a sample, a subset of individuals meant to represent the larger target population. The validity of any conclusion drawn from a study hinges on how well the sample mirrors the population it claims to represent.

The method of sampling directly determines external validity (generalizability). A well-chosen probability sample allows findings to be extended to the broader population. A poorly chosen or non-random sample introduces selection bias, meaning the study's results may only apply to the group studied and not to anyone else.

Exam questions typically present a research scenario and ask you to do one of three things: name the sampling method, identify the bias, or judge whether the study's conclusions can be generalized.

02Probability Sampling Methods

Probability sampling means every individual in the target population has a known, nonzero probability of being selected. This is the gold standard for generalizability.

Simple Random Sampling

How It Works: Every individual has an equal chance of selection (e.g., names drawn from a hat, random number generator)
Strengths: Eliminates selection bias; most straightforward
Weaknesses: Impractical for very large or geographically dispersed populations; may under-represent small subgroups by chance
Exam Keyword/Clue: "Randomly selected from a list of all patients"

Systematic Sampling

How It Works: Select every $k$ -th individual from an ordered list (e.g., every 5th patient on a hospital roster)
Strengths: Easy to implement; approximates randomness
Weaknesses: If the list has a hidden periodic pattern, it can introduce bias
Exam Keyword/Clue: "Every 10th patient on the clinic roster was chosen"

Stratified Sampling

How It Works: Divide the population into strata (subgroups based on a characteristic like age, sex, or disease severity), then randomly sample from each stratum
Strengths: Guarantees representation of important subgroups; increases precision for subgroup analyses
Weaknesses: Requires prior knowledge of the population's composition; more complex to execute
Exam Keyword/Clue: "Patients were divided by age group, then randomly selected from each group"

Cluster Sampling

How It Works: Divide the population into naturally occurring clusters (e.g., hospitals, schools, villages), randomly select entire clusters, then study everyone (or a random sample) within those clusters
Strengths: Cost-effective for large, geographically spread populations
Weaknesses: Less precise than simple random sampling; individuals within a cluster tend to be similar to each other, reducing variability captured
Exam Keyword/Clue: "Five hospitals were randomly chosen, and all patients in those hospitals were enrolled"

Multistage Sampling

How It Works: Combines methods in sequential stages (e.g., randomly select clusters, then stratify within clusters, then randomly sample individuals)
Strengths: Practical for national-level surveys; balances cost and representativeness
Weaknesses: Complex design; each stage can introduce its own source of error
Exam Keyword/Clue: "Counties were randomly selected, then clinics within those counties, then patients within those clinics"

Method	How It Works	Strengths	Weaknesses	Exam Keyword/Clue
Simple Random Sampling	Every individual has an equal chance of selection (e.g., names drawn from a hat, random number generator)	Eliminates selection bias; most straightforward	Impractical for very large or geographically dispersed populations; may under-represent small subgroups by chance	"Randomly selected from a list of all patients"
Systematic Sampling	Select every $k$ -th individual from an ordered list (e.g., every 5th patient on a hospital roster)	Easy to implement; approximates randomness	If the list has a hidden periodic pattern, it can introduce bias	"Every 10th patient on the clinic roster was chosen"
Stratified Sampling	Divide the population into strata (subgroups based on a characteristic like age, sex, or disease severity), then randomly sample from each stratum	Guarantees representation of important subgroups; increases precision for subgroup analyses	Requires prior knowledge of the population's composition; more complex to execute	"Patients were divided by age group, then randomly selected from each group"
Cluster Sampling	Divide the population into naturally occurring clusters (e.g., hospitals, schools, villages), randomly select entire clusters, then study everyone (or a random sample) within those clusters	Cost-effective for large, geographically spread populations	Less precise than simple random sampling; individuals within a cluster tend to be similar to each other, reducing variability captured	"Five hospitals were randomly chosen, and all patients in those hospitals were enrolled"
Multistage Sampling	Combines methods in sequential stages (e.g., randomly select clusters, then stratify within clusters, then randomly sample individuals)	Practical for national-level surveys; balances cost and representativeness	Complex design; each stage can introduce its own source of error	"Counties were randomly selected, then clinics within those counties, then patients within those clinics"

Stratified vs. Cluster sampling is the single most commonly confused pair.

In stratified sampling, you divide the population into subgroups and then take a random sample from every subgroup. The goal is to ensure each subgroup is represented.
In cluster sampling, you divide the population into groups and then randomly select entire groups. You do not sample from every cluster. The goal is logistical efficiency.

A useful memory anchor: in stratified sampling, you go into every stratum. In cluster sampling, you pick whole clusters and skip the rest entirely.

Systematic sampling and hidden periodicity: If a hospital roster alternates male and female patients and the investigator picks every 2nd name, the entire sample could end up being one sex. The vignette will hint at a pattern in the list to test whether you recognize this flaw.

03Non-Probability Sampling Methods

In non-probability sampling, selection is not random. Some individuals have no chance of being selected at all. These methods are easier and cheaper but inherently limit generalizability.

Convenience Sampling

How It Works: Enroll whoever is available or easy to reach (e.g., patients who happen to be in the waiting room)
Strengths: Fast, inexpensive, easy
Weaknesses: High risk of selection bias; poor generalizability
Exam Keyword/Clue: "Patients in the clinic that day were asked to participate"

Snowball Sampling

How It Works: Existing participants recruit future participants from their own networks
Strengths: Useful for hard-to-reach or stigmatized populations (e.g., IV drug users, undocumented individuals)
Weaknesses: Non-random; sample is shaped by social networks, not the true population
Exam Keyword/Clue: "Participants were asked to refer others they knew with the same condition"

Quota Sampling

How It Works: The researcher sets a target number for each subgroup and fills those quotas using non-random selection
Strengths: Ensures subgroup representation without requiring a full population list
Weaknesses: Selection within quotas is non-random, so bias persists
Exam Keyword/Clue: "The researcher enrolled patients until 50 males and 50 females were included"

Purposive (Judgmental) Sampling

How It Works: The researcher handpicks subjects based on their own judgment of who would be "informative"
Strengths: Useful in qualitative research or pilot studies
Weaknesses: Entirely dependent on researcher's judgment; highly prone to bias
Exam Keyword/Clue: "The investigator selected patients believed to be most representative"

Voluntary Response Sampling

How It Works: Participants self-select into the study (e.g., responding to an online survey or a call for volunteers)
Strengths: Easy to collect large samples
Weaknesses: Extreme selection bias: people with strong opinions or experiences are overrepresented
Exam Keyword/Clue: "A survey was posted online and respondents voluntarily completed it"

Method	How It Works	Strengths	Weaknesses	Exam Keyword/Clue
Convenience Sampling	Enroll whoever is available or easy to reach (e.g., patients who happen to be in the waiting room)	Fast, inexpensive, easy	High risk of selection bias; poor generalizability	"Patients in the clinic that day were asked to participate"
Snowball Sampling	Existing participants recruit future participants from their own networks	Useful for hard-to-reach or stigmatized populations (e.g., IV drug users, undocumented individuals)	Non-random; sample is shaped by social networks, not the true population	"Participants were asked to refer others they knew with the same condition"
Quota Sampling	The researcher sets a target number for each subgroup and fills those quotas using non-random selection	Ensures subgroup representation without requiring a full population list	Selection within quotas is non-random, so bias persists	"The researcher enrolled patients until 50 males and 50 females were included"
Purposive (Judgmental) Sampling	The researcher handpicks subjects based on their own judgment of who would be "informative"	Useful in qualitative research or pilot studies	Entirely dependent on researcher's judgment; highly prone to bias	"The investigator selected patients believed to be most representative"
Voluntary Response Sampling	Participants self-select into the study (e.g., responding to an online survey or a call for volunteers)	Easy to collect large samples	Extreme selection bias: people with strong opinions or experiences are overrepresented	"A survey was posted online and respondents voluntarily completed it"

Exam Application

When a vignette describes a non-probability method, the question usually asks about the study's limitation. The correct answer is almost always a statement about limited generalizability or selection bias. Even if the study's internal findings are valid, you cannot extend those findings to the general population because the sample was not randomly drawn.

04Sampling Bias and Its Relationship to Validity

Sampling bias (a form of selection bias) occurs when the method of selecting subjects causes the sample to differ systematically from the target population. This threatens external validity.

Sampling Bias

Definition: Systematic error due to a non-representative sample
Exam Relevance: The most frequently tested consequence of poor sampling

External Validity (Generalizability)

Definition: The degree to which study results apply beyond the study sample
Exam Relevance: Directly determined by sampling method

Internal Validity

Definition: Whether the study accurately measures what it intended within the study group
Exam Relevance: Less affected by sampling method; more affected by study design, blinding, confounders

Sampling Frame

Definition: The actual list from which the sample is drawn
Exam Relevance: If the frame is incomplete (e.g., only patients with insurance), bias is introduced even if random selection is used within that frame

Concept	Definition	Exam Relevance
Sampling Bias	Systematic error due to a non-representative sample	The most frequently tested consequence of poor sampling
External Validity (Generalizability)	The degree to which study results apply beyond the study sample	Directly determined by sampling method
Internal Validity	Whether the study accurately measures what it intended within the study group	Less affected by sampling method; more affected by study design, blinding, confounders
Sampling Frame	The actual list from which the sample is drawn	If the frame is incomplete (e.g., only patients with insurance), bias is introduced even if random selection is used within that frame

A critical exam concept: random sampling is not the same as random assignment (randomization).

Random sampling determines who enters the study from the population. It protects external validity.
Random assignment (randomization) determines which group a participant is placed into (e.g., treatment vs. control) after they are already enrolled. It protects internal validity by distributing confounders evenly.

Vignettes often describe a randomized controlled trial using convenience sampling. Students mistakenly assume the study is fully "random." The correct interpretation is that internal validity is maintained (due to randomization of groups) but external validity is limited (due to non-random sampling).

05Sample Size, Power, and Sampling Error

Though not a sampling "method," questions about sample size often appear alongside sampling concepts.

Sampling Error

Definition: The natural difference between a sample statistic and the true population parameter, purely due to chance
Exam Angle: Decreases as sample size increases

Standard Error (SE)

Definition: Quantifies sampling error; calculated as $SE = \frac{\sigma}{\sqrt{n}}$
Exam Angle: As $n$ increases, SE decreases, meaning the sample mean more precisely estimates the population mean

Power

Definition: The probability of correctly rejecting a false null hypothesis ( $1 - \beta$ )
Exam Angle: Increases with larger sample size

Confidence Interval Width

Definition: Range of values within which the true parameter likely falls
Exam Angle: Narrower with larger $n$

Term	Definition	Exam Angle
Sampling Error	The natural difference between a sample statistic and the true population parameter, purely due to chance	Decreases as sample size increases
Standard Error (SE)	Quantifies sampling error; calculated as $SE = \frac{\sigma}{\sqrt{n}}$	As $n$ increases, SE decreases, meaning the sample mean more precisely estimates the population mean
Power	The probability of correctly rejecting a false null hypothesis ( $1 - \beta$ )	Increases with larger sample size
Confidence Interval Width	Range of values within which the true parameter likely falls	Narrower with larger $n$

The exam tests this concept as follows: "A study found no significant difference between two treatments. Which of the following changes would most increase the likelihood of detecting a true difference?" The answer is increasing the sample size, because doing so increases statistical power and reduces sampling error.

06Comparison Table

"Patients were divided by ethnicity, then randomly selected from each group"

It Sounds Like...: Stratified sampling
But It Could Be Confused With...: Cluster sampling
The Key Discriminator: In stratified, you sample from every subgroup. In cluster, you select whole groups and skip others.

"Five community health centers were randomly selected, and all patients there were enrolled"

It Sounds Like...: Cluster sampling
But It Could Be Confused With...: Stratified sampling
The Key Discriminator: Entire clusters are chosen, not individuals from every group. The unit of randomization is the group, not the person.

"Every 7th name on the employee health record was selected"

It Sounds Like...: Systematic sampling
But It Could Be Confused With...: Simple random sampling
The Key Discriminator: The fixed interval ( $k = 7$ ) is the giveaway. True simple random sampling has no pattern.

"Participants were recruited by asking enrolled subjects to refer friends"

It Sounds Like...: Snowball sampling
But It Could Be Confused With...: Convenience sampling
The Key Discriminator: The referral chain is the hallmark. Convenience sampling has no referral; it just takes whoever is available.

"Patients who came to the ER during the study period were enrolled"

It Sounds Like...: Convenience sampling
But It Could Be Confused With...: Voluntary response sampling
The Key Discriminator: In convenience sampling, the researcher approaches available subjects. In voluntary response, subjects choose to participate on their own.

"An online survey was posted and anyone could respond"

It Sounds Like...: Voluntary response sampling
But It Could Be Confused With...: Convenience sampling
The Key Discriminator: Self-selection is the distinguishing feature. The researcher did not approach participants; they came on their own.

"The researcher selected 30 males and 30 females from the clinic"

It Sounds Like...: Quota sampling
But It Could Be Confused With...: Stratified sampling
The Key Discriminator: Quota sampling fills predefined slots without random selection. Stratified sampling randomly selects from each stratum.

If the Vignette Says...	It Sounds Like...	But It Could Be Confused With...	The Key Discriminator
"Patients were divided by ethnicity, then randomly selected from each group"	Stratified sampling	Cluster sampling	In stratified, you sample from every subgroup. In cluster, you select whole groups and skip others.
"Five community health centers were randomly selected, and all patients there were enrolled"	Cluster sampling	Stratified sampling	Entire clusters are chosen, not individuals from every group. The unit of randomization is the group, not the person.
"Every 7th name on the employee health record was selected"	Systematic sampling	Simple random sampling	The fixed interval ( $k = 7$ ) is the giveaway. True simple random sampling has no pattern.
"Participants were recruited by asking enrolled subjects to refer friends"	Snowball sampling	Convenience sampling	The referral chain is the hallmark. Convenience sampling has no referral; it just takes whoever is available.
"Patients who came to the ER during the study period were enrolled"	Convenience sampling	Voluntary response sampling	In convenience sampling, the researcher approaches available subjects. In voluntary response, subjects choose to participate on their own.
"An online survey was posted and anyone could respond"	Voluntary response sampling	Convenience sampling	Self-selection is the distinguishing feature. The researcher did not approach participants; they came on their own.
"The researcher selected 30 males and 30 females from the clinic"	Quota sampling	Stratified sampling	Quota sampling fills predefined slots without random selection. Stratified sampling randomly selects from each stratum.

07Traps and High-Yield Pearls

The most common trap on sampling questions is confusing stratified and cluster sampling. Both involve dividing a population into groups, but they diverge completely in how subjects are selected from those groups. If every group contributes subjects, it is stratified. If only selected groups contribute (and the rest are excluded), it is cluster.

A second frequent mistake is conflating random sampling with randomization. A randomized controlled trial that recruits from a single academic hospital has strong internal validity but weak external validity, because the sample may not reflect the broader patient population. Students see the word "randomized" and assume the study is free of all bias, which is incorrect.

Third, when a vignette describes volunteers or self-selected respondents, the tested concept is almost always voluntary response bias. These participants tend to have stronger feelings or more extreme experiences than the general population, which skews results.

Finally, remember that increasing sample size reduces sampling error and increases power, but it does not fix sampling bias. A very large convenience sample is still a biased sample. The only fix for sampling bias is improving the sampling method itself.

The core competency being tested across all sampling questions is your ability to evaluate a study's methodology: identify how subjects were selected, determine what bias that introduces, and judge whether the findings can be applied beyond the study population.

LevelUp. Archive