Sei sulla pagina 1di 19

Sampling is the process of selecting a subset of representative

individuals from a large group of individuals (Population/


Universe), with a view to drawing inferences about the larger
group on the basis of the information obtained from the subset.

Advantages

Time (Saves time)

Cost (Economical)

Accuracy (Sampling is more accurate than Census)

Destructive Nature
Sampling Error
Differences that exist between the results of the sample
and the true underlying population value
It is measurable
It decreases as the sample size increases

Non-Sampling Errors
All the errors except sampling error
Simply includes all the aspects of the research process
where mistakes and deliberate deceptions can occur
It is not easily measurable
It increases as the sample size increases
It puts a bias in the results of unknown direction &
magnitude
Unfortunately these mistakes and deceptions occur in
great frequency
Non-Sampling Errors (Examples)
Faulty Problem Definition
Defective Population Definition
Frame is Non-representative of the Population
Non-response Error
Measurement Error
Improper Causal Inferences
Poor questionnaire design
Auspices Bias: Error because of the respondents
perception of who is conducting the survey. Peoples true
feelings may not come out when they know the source of
the study.
Interviewer-related Errors
Non-Sampling Errors (Examples)
Data Processing Errors
Coding the responses, recording the codes, entry of codes in
computer
Data Analysis Error
Simple mistakes can occur in data analysis, such as adding
numbers incorrectly. Most errors are more complex that this.
Interpretation Errors
People can interpret a set of data to suit their own purposes.
This bias may be explicitly deliberate or subconscious.
Non Probability Sampling
Is a sample drawn from a population in which there is no
known specified probability for each element to be
included in the sample.
Probability Sampling
Is a sample which is drawn in such a way that every
element in the population has a known non-zero
probability of being included in the sample.
The likelihood for a sample to be representative of the
parent population is higher because of its unbiased
character
Convenient/ Accidental
Chosen from anyone in sight in a street, market place, bus
station, etc.
Judgmental/ Purposive
Uses the experience to select the elements that best will
best serve the purpose of the research.
Quota
Based on the prior knowledge about the population the
quotas for various segments of the population are
specified.
Snow Ball
Hardly anything is known about the population. From
the indication of very few picked up samples, the
sampling procedures moves further, and likewise goes
on.
Simple Random Sampling
Every element in the population has an equal chance of being
included
Systematic Sampling
Well organized population
Stratified Sampling
Population is divided into a number of mutually exclusive non-
overlapping groups (strata) which are homogenous within the strata
but heterogeneous among themselves and samples are selected from
those groups
Cluster Sampling
Clusters are groups which are heterogeneous within the strata but
homogenous among themselves and samples are selected from some
of these groups
Element: An unit of a population about which information
or data are to be collected in a particular research.
Population: Aggregate of all the elements of the research
topic/ Area.
Sampling Unit: An element or a collection of a number of
elements (Multistage Sampling).
Sampling Frame: Actual list of all sampling units.
Multistage Sampling: Selecting a sample in different stages,
e.g., to draw a national sample of households at three stages
First stage: Select few provinces from all provinces (Primary
sampling unit)
Second stage: Select few districts from all districts (Secondary
sampling unit)
Third stage: Select few households from all households (Ultimate
sampling unit)
We have noted that if is known, the confidence interval estimate for the
population mean is obtained from equation
N n
X z / 2
n N 1
N n
X z / 2
n N 1

Here, ( X ) : the sampling error (or precision) ,
z / 2 : confidence level (or reliabilit y), and
N n
: the standard error
n N 1
Hence, precision (d) confidence level (or reliabilit y) standard error

Note that when sampling without replacement from finite populations,


the finite population correction factor (fpc) served to reduce the standard
error by a factor equal to (N-n)/(N-1).
Now, from the above equation we can derive the sample size as follows:

Given,
N n
d z / 2
n N 1
z 2 2 N n
d 2
, [for practical purposes we can use N for (N - 1)]
n N
d2N N n d2N N d2N N z 2 2 d 2 N N
2 2 2 2 1 1 2 2
z n z n z n z
2 2
n
Nz 2 2
n 2 2
z d2N
Please note that to calculate the sample size normally four factors must be known:

Population size (N): The actual population of the universe.


The desired significance level (): The z value can be determined once the desired
level of significance is known.
Precision (d): The accuracy or sampling error (plus or minus) that we are willing to
accept in using the sample statistic to estimate the population parameter.
Population standard deviation (): An estimate of the standard deviation must be
available in order to determine the required sample size. In some cases the standard
deviation of the variable is known. In other instances, past (historical) data may be
available that can be extrapolated to estimate the current standard deviation. If the
standard deviation cannot be estimated from past data, a pilot study can be taken,
and results can be used to obtain an estimate of the standard deviation.

Example

The controller of a bank would like to determine the sample size to estimate the average
monthly amount spent by the credit card holders. The controller would like to estimate
the population mean to be within $5 of the true value. He would like to be 95%
confident of correctly estimating the true mean. Based upon previous studies
undertaken, the standard deviation is estimated as $21. The number of total credit card
holders are 10,245.
In many cases, instead of using absolute precision we prefer to use
relative precision figures in terms of mean (e.g., 5%, 10%, etc). In such
cases the relative precision (do) is defined as:
d
do
X

Using, d d o X and coefficien t of variation , c o c o X , the sample size,
X
Nz 2 2
2
Nz Nz
2
2 2 2 2
X Nz 2 co
n 2 2
z d 2N 2 2
z 2 co d o N
2 2
z do X N z do X N
2 2 2 2 2 2

2
X
Example
A researcher wishes to determine the sample size to determine the IQ level
of graduate level business students in a business school. The total number
of graduate students in that school is 867. A pilot survey of 16 students
shows the mean IQ is 124 with a standard deviation of 14. Determine the
sample size at a confidence level of 95% with a precision of 3%.
In many cases when the population is infinite (Very large), we can avoid
the use of the population correction factor for calculating the standard
error (as the value will approximate to one). In that case the sample size
determination formula will be:

z 2 2 z 2 2
dz d 2
n 2
, for absolute precision
n n d
z 2 2 z 2 2
2
z 2 co
n 2
2
2
, for relative precision
d
2 d0
d0 X
Sample Size Determination for the Proportion
The methods of sample size determination that are utilized in estimating a
true proportion are similar to those employed in estimating a mean.

p(1 p) N n
ps z / 2 , [ ps is the sample proportion ]
n N 1
p(1 p) N n
ps z / 2
n N 1
Here, ( - p s ) : the sampling error (or precision) ,
z / 2 : is the confidence level, &
p(1 p) N n
: the standard error
n N 1
Hence, precision (d o ) confidence level standard error

Note that in the above formula, precision (do) is always relative as it is


given in percentages. Now, from the above formula we can derive the
sample size as follows:
n = (Nz2pq)/(Ndo2 + z2pq) , for finite population
= (z2pq)/do2 , for infinite population
where,
n: Sample Size
N: Population size
z: Reliability (depends on level of significance)
do: Precision
p: proportion (0.5 for maximum variance)
q: 1 p

Example

A researcher is conducting a research to find out the true proportion of


students satisfied with the student cafeteria facilities to be correct within 9%.
Based upon a similar study undertaken about a year back, the true proportion
satisfied with the dining facilities is estimated to be 0.25. With this
information, determine the sample size at a confidence level of 90%.
Budget and Sample Size

Budget plays an important role in sample size determination. In many cases it


so happens that we have to reduce the sample size due to budget constraint.
In such cases we can justify our reduction of sample size by changing either
the precision level or level of significance. The following example will explain
the sample-budget relationship.

Example

A Mobile phone company wants to estimate what proportion of the


university students use mobile phone. For the survey they set a budget of
$1000. Each sample survey costs $1.8 per sample. The survey has a capital cost
involvement of $400. The company likes the result to be accurate within 5%.
They also suggests a confidence level of 95%. The study was conducted in a
university with a student of about 4,424. Find the sample size.
Here, d 0.05, p 0.5( to be in the safer side), q 0.5, z 1.96 (at 0.05), N 4,424
o
Nz 2 pq 4,424 1.96 2 (0.5 0.5) 4248.8096
Hence, n
z 2 pq d N 1.96 2 (0.5 0.5) 0.05 2 4,424 0.9604 11.06
2
o
353 .36 354
The researcher should take a sample size of 354

But the budget allows a sample size of only (1000 400)/1.8 334

Hence the sample size should be 334. This sample size can be justified with a significance
level of

Nz 2 pq 4,424 z 2 (0.5 0.5) 1106 z 2


Here, n 334
2
z pq d N
2 2 2
z (0.5 0.5) 0.05 4,424 0.25 z 2
11.06
o
3.31z 2 0.25 z 2 11.06 3.06 z 2 11.06 z 2 3.614 z 1.901
This z value will justify an 0.0562

Potrebbero piacerti anche