Probability Distributions and Sampling Distribution (part 2: 2.5–2.7)
Open in the Padhaiko app2.5 Population and sample
Statistical inference bridges the gap between a small observed dataset and the wider system from which it was drawn. Understanding the distinction between a population and a sample—and how the sample is selected—is essential before applying any probability distribution to real data.
| Aspect | Population | Sample |
|---|---|---|
| Definition | The complete set of all items, measurements, or individuals of interest | A finite subset of elements selected from the population |
| Size Notation | (finite population) or (infinite/conceptual process) | () |
| Summary Quantities | Parameters (fixed, usually unknown constants: ) | Statistics (random variables computed from sample data: ) |
| Data Collection | Census (expensive, destructive, or impossible for ongoing processes) | Sampling (fast, economical, feasible even for destructive testing) |
2.5.1 Sampling Techniques and Errors
A sample must be representative of the population to avoid systematic distortion (bias). Sampling methods fall into two broad categories:
| Category | Method | Mechanism and Engineering Use Case |
|---|---|---|
| Probability Sampling (known nonzero selection probability) | Simple Random Sampling (SRS) | Every subset of size has equal chance of selection; implemented with replacement (SRSWR) or without replacement (SRSWOR) |
| Probability Sampling | Stratified Sampling | Population is partitioned into homogeneous strata (e.g., shifts or suppliers), and SRS is drawn proportionally from each stratum |
| Probability Sampling | Systematic Sampling | Select a random start and then every -th item on a production line () |
| Probability Sampling | Cluster Sampling | Divide population into natural groups (e.g., boxes/batches), randomly pick clusters, and inspect all items inside selected clusters |
| Non-Probability Sampling | Convenience / Purposive / Quota | Items selected by ease of access or expert judgment; cheap for exploratory checks, but cannot support valid probability error bounds |
- Sampling Error: Natural chance variation between a sample statistic () and the true population parameter (); decreases as sample size grows.
- Non-Sampling Error: Mistakes from biased selection, miscalibrated sensors, non-response, or recording errors; cannot be cured simply by increasing .
2.4
A finite population consists of resistor tolerance deviations .
- Compute the population mean and population variance .
- List all possible samples of size drawn without replacement (SRSWOR) and construct the sampling distribution of .
- Verify that and .
Population parameters ():
Enumerate all SRSWOR samples of size :
| Sample | ||||||
|---|---|---|---|---|---|---|
| Sample Mean |
Each sample has probability , so the sampling distribution of is:
Verify mean and variance of :
Using the finite-population formula:
, ; and , matching .
2.6 Sampling distribution of mean and proportion
A sampling distribution is the probability distribution of a sample statistic (such as or ) across all possible random samples of fixed size drawn from the same population. The standard deviation of a sampling distribution is called its standard error ().
2.6.1 Sampling Distribution of the Sample Mean ()
Mean, Variance, and Distribution of the Sample Mean
Let be a random sample from a population with mean and variance :
- Infinite population or sampling with replacement (SRSWR):
- Finite population of size sampled without replacement (SRSWOR):
where is the Finite Population Correction (FPC) factor. (In practice, FPC is applied whenever the sampling fraction ; otherwise .)
- Exact distribution for a Normal population :
- Known : , so for any sample size .
- Unknown (estimated by sample SD ): (Student's -distribution with degrees of freedom).
2.6.2 Sampling Distribution of the Sample Proportion ()
When each sampled unit has a binary attribute (e.g., defective/non-defective) with population proportion , let be the count of successes in a random sample of size . The sample proportion is .
Mean, Standard Error, and Normal Approximation of Sample Proportion
For a random sample of size from a population with true proportion ():
When and , is approximately normally distributed:
2.5
A large production run of SMD resistors has a true defect proportion of . A quality inspector draws a random sample of resistors.
- Find the mean and standard error of the sample proportion .
- Find the approximate probability that more than of the sampled resistors are defective ().
Check normal-approximation conditions and compute and : Since and , the normal approximation applies:
Standardize and evaluate :
(Without rounding , .)
- , ; 2. ().
2.7 Central limit theorem
The Central Limit Theorem (CLT) explains why the Normal distribution dominates engineering statistics: even when individual measurements follow a skewed, uniform, or discrete distribution, the average (or sum) of enough independent observations becomes approximately Normal.
Central Limit Theorem (Lindberg–Lévy)
Let be independent and identically distributed (i.i.d.) random variables from any population distribution with finite mean and finite variance . As the sample size becomes large (), the standardized sample mean converges in distribution to the standard normal:
Equivalently, for sufficiently large :
2.7.1 Practical Guidelines and Engineering Significance
- Rule of Thumb (): For most unimodal or moderately skewed distributions, gives an accurate normal approximation for . If the underlying population is already symmetric, often suffices; if heavily skewed (e.g., exponential), may be needed. If the population itself is Normal, the distribution of is exact for every .
- Why Engineering Depends on the CLT:
- Unknown Population Shape: Engineers rarely know the exact PDF of packet delays, concrete strengths, or solder joint resistances; the CLT allows -intervals and hypothesis tests using only and (or for large ).
- Sum of Independent Errors / Tolerances: Cumulative manufacturing tolerances and thermal noise across many stages naturally follow a Normal curve.
- Bridge to Discrete Models: Binomial (sum of Bernoulli trials) and Poisson (large ) converge to Normal distributions via the CLT.
2.6
The mass of individual machined aluminum brackets has a non-normal distribution with mean and standard deviation . A shipping crate holds a random sample of brackets. Using the Central Limit Theorem:
- Find the probability that the sample mean mass lies between and (, ).
- Find the probability that the total mass of the brackets exceeds ().
Apply the CLT (): For the sample mean :
So .
Part 1 — Compute :
Part 2 — Compute : The total mass has mean and standard deviation :
(Using exact gives .)
- ; 2. ( with -decimal ).
Practice Exercises — Sampling Distributions & Central Limit Theorem
Distinguish between a parameter and a statistic, and between sampling error and non-sampling error. Explain when the Finite Population Correction (FPC) factor is required and why it disappears as .
Parameter vs. Statistic & Errors:
- A parameter is a fixed numerical summary of an entire population (), whereas a statistic is a random variable computed from a sample () whose value varies across samples.
- Sampling error is the inherent chance difference arising from observing only a subset; non-sampling error arises from bias, instrument error, or flawed data recording.
Finite Population Correction (FPC): When sampling without replacement from a finite population of size , . The FPC matters when (sampling more than of the population). As for fixed , , recovering the infinite-population standard error .
Parameters describe populations while statistics describe samples; sampling error is chance variation from partial observation while non-sampling error comes from bias or measurement flaws. FPC is used for SRSWOR when and approaches as .
A finite population consists of the values . Samples of size are drawn at random.
- Compute the population mean and variance .
- Find the mean and standard error if sampling is done without replacement.
- Find if sampling is done with replacement.
Population parameters ():
Without replacement (SRSWOR, ):
With replacement (SRSWR, ):
- , ; 2. Without replacement: , ; 3. With replacement: .
A warehouse holds battery packs whose capacities have mean and standard deviation . A sample of packs is drawn without replacement.
- Determine using the finite population correction factor.
- Find the approximate probability that the sample mean exceeds ().
Check sampling fraction and compute : Since , apply the FPC:
Apply CLT to evaluate :
(Exact gives .)
- ; 2. ( with ).
In a large campus network, of traffic packets use IPv6 (). If a packet sniffer captures a random sample of packets, find the standard error of and the probability that the sample proportion lies between and ().
Verify normal condition and compute : and :
Standardize and find :
and .
The lifetime of an industrial sensor battery has mean and standard deviation (distribution shape unknown).
- For a random sample of batteries, use the Central Limit Theorem to find (given ).
- What minimum sample size is required so that (given )?
Part 1 — Sample mean probability for :
Part 2 — Minimum sample size for margin of error : We require , so :
Rounding up to the next integer gives .
- ; 2. Minimum sample size .