Sampling PDF

UNIT 16 BASIC CONCEPTS OF SAMPLING Structure 16.0 Objectives 16.1 Introduction 16.2 Census and Sample Survey 16.2.1 Population and Census 16.2.2 Sample and Sample S w e y 16.3 Some Concepts 16.3.1 Parameter 16.3.2 Statistic 16.3.3 Estimator and Est...

UNIT 16 BASIC CONCEPTS OF SAMPLING Structure 16.0 Objectives 16.1 Introduction 16.2 Census and Sample Survey 16.2.1 Population and Census 16.2.2 Sample and Sample S w e y 16.3 Some Concepts 16.3.1 Parameter 16.3.2 Statistic 16.3.3 Estimator and Estimate 16.4 Non-Sampling and.Sampling Errors 16.4.1 Non-Sampling Error 16.4.2 Sampling Error *' 16.5 Advantages of Sample Survey 16.6 Typ& of Sampling 16.6.1 Probability Sampling 16.6.2 Non-Probability Sampling 16.6.3 Mixed Sampling 16.7 Sampling Distribution 16.8 Standard Error of a Statistic 16.9 Desirable Properties of an Estimator 16.9.1 Unbiasedness 16.9.2 Minimum Variance 16.9.3 Consistency and Efficiency 16.10 LetUs SurnUp 16.11 Keywords 16.12 Some Usehl Books 16.13 Answers1 Hints to Check Your Progress Exercises 16.0 OBJECTIVES After going through this unit you should be able to: explain the concepts of population, sample, parameter, statistic, estimator and , estimate; distinguish between a census and a sample survey; explain the advantages of a sample survey; distinguish between sampling error and non-sampling mr,. 'explain the concept of sampling distribution; and explain the concept of standard error. Sampling Theory and S u r v e y Techniques 1 6 1 INTRODUCTION We need data for the construction of national income accounts, input-output tables, various production indices, price indices and a host of other quantitative indicators. It is very clear that without the relevant data, we will not be able to formulate policy objectives for a complex economy like ours. In a sense, modem society is increasinglybecoming an information society. In this society, various economic and social prkesses are represented by certain quantitative characteristics that require various kinds of information in the form of data. The task oficollecting data is getting increasingly complex and difficult. The total numberof units to be consulted and investigated for the required information may be too large and our resources in terms of money, time or personnel may be limited Moreover, obtaining error-he information from such a large-scale investigation makes the job even more daunting. As a result, very often we try to obtain the required information fkom a smaller group that is easier to handle and control.Here, however, it is important to ensure that this smaller group is truly representative of the entire c )llection of relevant units. The subject matter of sampling provides a mathematical theory for obtaining such kind of a representative group. 16.2 CENSUS AND SAMPLE SURVEY In this Section, we will distinguish between the census and sampling methods of collecting data. We will try to explain the meaning and covemge of census survey and sample survey. 16.2.1 Population and Census We have a collection of units relevant for a particular enquiry. A unit, in this connection, is an entity on which we can make observations according to a well- defined procedure. The entire collection of such units is called apopulation or universe. Thus,we may have a population of human beings, cattle, trees, prices, production, etc. You can make out that a population can be finite or infinite. If the number of units is finite, it is a finite population and if the number of units is infinite, it is an example of an infinite population. Usually in practice, we are concerned with a finite population. When an inquiry is based upon obtaining information from all the units of a pdpulation, the procedure is known as the complete enumeration method or the census method. 16.2.2 Sample and Sample Suwey When we have a collection of a part or section of the population, it is called a sample. A census, as we have seen earlier, is based upon obtaining information h m every member of the population. However, in order to obtain information about certain characteristic of the population, we need not always resort to a census. In practice, we get quite satisfactory results by studying an appropriate sample from the population. The procedure of obtaining a sample is known as sample survey. In the case.of a census, we examine the entire population; on the other hand, when we take a sample, we consider a representative fraction of the nnniilatinn and iicp the camnl~ infnrmatinn tn infer ahniit t h entire ~ nnniilatinn  Basic Concepts of 16.3 SOME CONCEPTS Sampling We explain below some of the concepts frequently used in sampling theory. 16.3.1 Parameter In a statiitical inquiry, our interest lies in one or more characteristics of the population. A measure of such a characteristic is called aparameter. For example, we may be interested in the mean income of the people of some region for a particular year. We may also like to know the standard deviation of these incomes of the people. Here, both mean and standard deviation are parameters. Parameters are conventioilally denoted by Greek alphabets. For example, the population mean can be denoted by p and population standard deviation can be i denoted by o. It is important to note that the value of a parameter is computed from all the populatioil observations. Thus, the parameter 'mean income' is calculated from all the income figures of different individuals that constitute the population. Similarly, for the calculation of the parameter 'correlation coefficient of heights and weights', we require the values of all the pairs of heights and weights in a population. Thus, we can define a parameter as afitnction of the population values. If 8 is a parameter that we want to obtain from the population values XI,X, ,.X, ,then 16.3.2 Statistic While discussing the census and the sample survey, we have seen that due to various constraints, sometimes it is difficult to obtain information about the whole population In other words, it may not be always possible to compute a population parameter. In such situations, we try to get some idea about the parameter fiom the information obtained from a sample drawn fiom the population. This sample information is sumrnarised in the form of a stati.vtic. For example, sample mean or sample median or sample mode is called a statistic. Thus, a statistic is calculated fiom the values of the units that are included in the sample. So, a statistic can be defined as u function of the sample values. Conventionally, a setistic is denoted by an English alphabet. For example, the sample mean may be denoted by 2 and the sample standard deviation may be denoted by s. If T is a statistic that we want to obtain from the sample values x, ,x, ,. xn,then 16.3.3 Estimator and Estimate The basic purpose of a statistic is to estimate some population parameter. The procedure followed or the formula used to compute a statistic is called an estimutor and the value of a statistic so computed is known as an estimate.. XI +x2 +. a * + xn - --1 " If we use the formula Z = xi for calculating a statistic,then n n i=l this formula is an estimator. Next, if we use this formula and get jj = 10, @en this ' 10' is an estimate. Sampling Theory and SU-rrry Techniques 16.4 NON-SAMPLING AND SAMPLING ERRORS As mentioned above the basic purpose of sampling is to draw inferences about the population on the basis of the sample. For example, we have to find out the per capita income of a village. Due to shortage of time, money and personnel we do not undertake a complete census and opt for a sample survey. In this case it is very likely that the per capita income obtained fiom the sample is not equal to the actual per capita income of the village. This discrepancy could arise because of two reasons: i) Since we are collecting data fiom only a part of the population (i.e., the sample selected by us), sample mean (per capita income in this case) is not equal to population mean. If at all both are equal; it is a rare coincidence! If we take sample mean as population mean we are committing an error called sampling enor. ni A second source of error could arise because of wrong reporting or recording or tabulation or processing of data. This type of error is termed non-sampling error. Remember that non-sampling error, as its name suggests, has nothing to do with our sampling process. Wrong reporting or recording or processing of data can take place in a sample survey also. We explain the sources of these errors below. 16.4.1 Non-Sampling Error Various sources of non-sampling error are given below: 1) Error due to measurement It is a well-known fact that precise measurement of any magnitude is not possible. If some individuals, for example, are asked to measure the length of a particular piece of cloth independently up to, say, two decimal points; we can be quite sure that their answers will not be the same. In kt,the measuring instrument itself may. not have the same degree of accuracy. In the context of sampling the respondents of an inquiry, for example, may not be able to provide the accurate data about their incomes. This may not be a problem with individuals earning fixed incomes in the form of wages and salaries. However, self-employed persons may not be able to do so. 2) Error due to non-response Sometimes the required data are collected by mailing questionnaires to the respondents. M & I ~of such respondents may return the questionnaires with incomplete answers or may not return them at all. This kind of an attitude may be due to: a) the respondents are too casual to fill up the answers to the questions asked b) they are not in a position to understand the questions, or c) they may not like to disclose the information that has been sought. We should note that the error due to non-response may also arise because of the possibility of the questionnaire being lost in transit. If the data are collected through personal interviews, some of the reasons for the error due to non-response pointed out above may not arise. However, in that case this error may arise because some of the individuals:  a) may not like to give the information, or Baslc Concepts of Sampling b) may not simply be available even after repeated visits. 3) Error in recording This type of error may arise at the stage when the investigator records the answers or even at the tabulation stage. A major reason for such error is the carelessness on the part of the investigator. 4) Error due to inherent bias of the investigator Every individual suffers b m personal prejudices and biases. Despite the provision of the best possible training to the investigators, their personal biases may come into play when they interpret the questions to be put to the respondents or record the answers to these questions. In complete enumeration the extent of non-sampling error tends to be significantly large because, generally, a large number of individuals are involved in the data collection process. We try to minimise this error through: i) a careful planning of the survey, ii) providing proper training to the investigators, iii) making the questionnaire simple. However, we would Like to emphasize that complete enumeration is always prone to large non-sampling errors. 16.4.2 Sampling Error By now it should be clear that in the sampling method also, non-sampling error may be committed. It is almost impossible to make the data absolutely fi-eeof such errors. However, since the number of respondents in a sample survey is much smaller than in census, the non-sampling error is generally less pronounced in the sampling method. Besides the non-sampling errors, there is sampling error in a sample survey. Sampling error is the absolute difference between the parameter and the corresponding statistic, that is, IT -01. Sampling error is not due to any lapse on the part of the respondent or the investigator or some such.reason. It arises because of the very nature of the procedure. It can never be completely eliminated. However, we have well developed sampling theories with the help of which the effect of sampling m r can be minirnised. 15.5 ADVANTAGES OF SAMPLE SURVEY There are important advantages of a sample survey over complete enumeration - or census method. Some of these advantages are mentioned below. i) Practicability Sometimes, a census may not be practicable due to the enormity of the task requira in the collection of data of a large population In such a situation, a sample survey may be quite practicable. ii) Speed -1 >--- - _. L - _ _ 1 1 _ - ~ -_ 3. 3 _ 3 _ A - _ __ 1. - 1. - - Sampling Theory and census. This may be an important advantage, particularly, when the information Survey Techniques is urgently needed. iii) Accuracy In any survey, census or sample, the required information is obtained by filling in the questionnaires.It has been observed that more accurate results are achieved when the investigato~~ themselves fillin the questionnaire instead of the respondents filling it. Again, personal interviews may result in more accurate information than sending the questionnairesto the respondents by post and requesting them to fill in these questionnaire. Normally, the umber of investigators involved in an inquiry varies directly with the number of r & ndents covered in the inquiry. As a result, personal interviews prove to be easier in the case of a sample survey than in a census. In fact, a sample survey has a greater scope to employ more efficient and better-trained investigators. In the case of a sample survey, the investigators can devote more time to each respondent. Thus, although a sample survey can have less coverage than a census, it may have greatef accuracy of the results. iv) Cost It is obvious that a sample survey results in less expenditure than a complete enumeration. After all, in a survey only part of the population is involved. The cost components of an inquiry are: a) Overhead cost of the organisation conducting the survey, b) Cost of collecting the data, c) Cost of processing and tabulating the data, and d) Cost of publication of results of the survey. In this cost break-up, items (b) and (c) are in the nature of variable costs, whereas (a) and (d) are the fixed cost items. As a result, items (b) and (c) will definitely be much smaller in a sample survey than for a census. We should note that the designing of a proper sample survey and the selection of an appropriate sample may entail considerable expenditure. However, generally it has been observed *$at a sample survey is less costly than complete enumeration. Check Your Progress 1 1) Define the following concepts: a) Population b) Sample,-: c) paramGter d) Statistic 2) Distinguish between the following: a) Estimator ahd Estimate b) Census and Sample Survey rl Samnlinv errnr and nnn-samnlinn error  Basic Concept8 of Sampling 3) What are the advantages of sampling over a census?...................................................................................................................................................................................................................................... 16.6 TYPES OF SAMPLING The method of selecting a sample fiom a given population is called sampling. Basically there are two types of sampling, viz., probability sampling and non- probability sampling. In probability sampling the sampling units are selected according to some chance mechanism or probability of selection. On the other hand, non-probability sampling is based on judgement or discretion of the person making a choice. Thus in non-probability sampling certain units may be selected because of convenience or they serve a purpose or the researcher feels that these units are representative of the No random selection on the basis of chance mechanism is involved here.' 16.6.1 Probability Sampling It is-alsocalled random sampling. It is a procedure in which every member of the population has a chance or probability of being selected in the sample. It is in this probabilistic sense that the sample is random. The word 'random' does not mean that the sample is obtained in a haphazard manner without following any rule. Random sampling is based on the well-established principles of probability theory. There are quite a few variants of the random sampling, viz., simple random sampling, systematic random samphng and stratified random sampling. We discuss these types below. a) Simple Random Sampling If there is not much variation in the characteristics of the members of a population, we can follow the method of simple random sampling. In this method, we umsider the population in its entirety as a homogeneous group and follow the principle of random sampling to choose the members for the sample. There are two variants of simple random sdlnpling, viz., simple random sampling with replacement (SRSWR) and simple random sampling without r e p l a c k t (SRSWOR). This difference pertains to the way the sample units are selected. According to the procedure of simple random sampling with replacement (SRSWR), we draw one unit h m the population, note down its features and put it back to the whole lot in the sense that the unit again becomes eligible fbr selection. In this way, the total number of units in the population always remains the same. In other words, the composition of the population remain9 unchanged, and each member of the population has the same chance or probability of being selected  1 in the sample. In fact, if N is the size of the population, this probability is - N' On the other hand, in the case of simple random sampling without replacement, the unit once selected is not returned to the population in the sense that it becuines ineligible for selection again. As a result, after each d m i v e draw, the composition of the population changes. Therefore, for subsequent draw fiom the population the probability of any particular unit being picked up also gets changed. Let us try to understand this. Suppose, the population size is Nand we want to draw a sample of size n h m it by the principle of SRSWOR. Before the first unit is 1 drawn, each unit of the population has the same chance (-) of being selected N in the sample. Once the first member of the sample is selected, each of the 1 remaining N-1 members of the population has an equal chance of -of N-1 selection in the sample. Finally, before the nthmember of the sample is chosen, each of the remaining members of the population has an equal chance of 1 -. - 1, of being included in the sample. N-(n+1) N-ntl We should note that fiom a population of size N, the number of samples of size n that can be drawn with replacement is N~ and the number of samples that can be drawn withailt replacement is c,. Example 16.1 Suppose a population consists of the following 5 units (4,5,7,9, 10). How many samples of size 2 can be drawn from it? 0 If we follow the procedure of SRSWR the numb& of samples that can be selected is = N n =52 = 2 5. The possible samples are given by. (4, 41, (4,519 (4,7), (4, 91, (4, 101, (5, 41, (5, 51, (5, 71, (5, 91, (5, lo), (7, 4), (7, 51, (7, 7), (7, 91, (7, 101, (9, 4), (9, 51, (9, 71, (9, 91, (9, 101, (10, 41, (10, 51, (10, 71, (10, 91, (10, 10). We should note that in sampling with &lacement, the order in which the units are selected also matters. Thus, (4, 10) and (10,4) are considered as two different samples. ii) If we follow the procedure of SRSWOR the number of samples that can be selected is The possible samples are given by (4, 51, (5,7), (7, %, (9, 101, (4, 71, (4,919 (4, 101, (5,913 (5, 101, (7, 10). We should note that in sampling without replacement, once a member is selected, it cannot be-selectedagain. Thus, samples like (4,4), (5, 5) etc. cannot be selected. Similarly, if a sample like (4,5) is selected, then another camnle like I S 4) cannot he selected.  - - - - li b) Systematic Random Sampling In this variant of random sampling, only the first unit of the sample is selected at random fi-om the population. The subsequent units are then selected by following Boric Concepts of Sampling some definite rule. For example, suppose, we have to choose a sample of I 1 agricultural plots. In systematic random sampling, we begin with selecting one plot at mndom and then every 10" plot may be selected. I c) Stratified Random Sampling Stratified random sampling is the appropriate method if the population under I consideration consists of heterogeneous units. Here, first we divide the population into certain homogeneous groups or strata. Secondly, fbm each stratum some units are selected by simple random samphg. Thirdly, after selecting the units h m each stratum, they are mixed'together to obtain the final sample. Let us consider an example. Suppose, we want to estimate the per capita income of Delhi by a sample survey. It is common knowledge that Delhi is characterised by rich localities, middle class localities and poor localities in terms of the income groups of the people living in these localities. Now, each of these different localities can constitute a stratum from which some people may be selected by adopting simple random sampling procedure. d) Multi-Stage Random Sampling Let us consider a situation where we want to obtain information b m a sample of households in a large city, say, Delhi. Sometimes, it may not be possible to directly take a sample of households because a list of all the households may not be easily obtained. In such a situation, one may resort to take samples in various stages. Generally, the city is divided into certain geographical areas for administrative purposes. These areas may be termed as city blocks. So in the first stage, some of such blocks may be selected by random sampling. In the next stage, fi-om each of the selected blocks in the first stage, some households may be selected again by the principle of random sampling. In this way, ultimately a sample of households from a large city may be obtained. The above-mentioned example is the case of a two-stage random sampling. However, if the nature of the inquiry so demands, the method of sampling can be extended to more than two stages. 16.6.2 Non-Pro ba bility Sampling We have considered the method of random sampling and some of its variants above. It should be clear that the basic objective of the principle of random sampling is to eliminate or at least rninimise the effect of the subjective bias of the investigator in the selection of the population sample. But for certain purposes, there is a need for using-discretion. For example, suppose a teacher has to choose 4 participants fi-om a class of 30 students in a debate competition. Here, t.teacher may select the top 4 debaters on the basis of her o& conscious judgement about the top debaters in the class. This is an example of purposive sampling. In this method, the purpose of the sample guides the choice of certain members or units of the population. 16.6.3 Mixed Sampling In mixed sampling, we have some features of both non-probability sampling and random samphg. Suppose, an institute has to send 5 students for managerial training in a company during the summer vacation. Initially; it may shortlist about 20 students who are considered to be suitable for the training by applying its own discretion. Then from these 20 students, 5 students may finally be selected by random sampling. Sampling Theory and Survey Techniques 16.7 SAMPLING DISTRIBUTIUON By now it should be clear that generally the size of a sample is much smaller than the parent population. Conseguently, many samples can be selected h m the same population which are'different h m one another. Since an estimate of a parameter depends upon the sample values, and these values may change from one sample to another, there can be different estimates or values of a statistic for the same parameter. This variation in values is called sampling.jZuctuation. Suppose, a number of samples, each of size n, are drawn from a population of size N and for each sample, the value of the statistic is computed. If the number of samples is large, these values can be ammged in the form of a relative fkquency distribution. When the number of samples tends to infinity, the resultant relative frequency distribution of the values of a statistic is called the sampling distribution of the given statistic. Suppose, we are interested in estimating the population mean (which is a parameter), denoted by p. A random sample of size n is drawn h m this population 1 (of size N). The sample mean = -b, is a statistic corresponding to the n population mean p. We should note that is a random variable as its value changes from one sample'to another in a probabilistic manner. Example 16.2 Consider a population consisting of the following 5 units: 2,4, 6, 8, and 10. Suppose, a sample of size 2 is to be selected from it by the method of simple random sampling without replacement. We want to obtain the samphg distribution of the sample mean and its standard error. The number of samples that can be selected without replacement. The possible samples along with the corresponding sample means (F) are presented in Table 16.1. Table 16.1 : Possible Samples and Sample Means I Sample Sample Mean ( 2 ) Now, we can have a frequency distribution of the sample means:  Table 16.2: Frequency Distribution of Sample Means Basic Con~ept8of Sampiing Sample Mean Frequency From the frequency distribution given in Table 16.2, we can present the probability distribution of the sample mean as given in Table 16.3. Table 16.3: Sampling Distribution of Sample Means Sample Mean (x) Probability [&) We note here that C.f ,whlch, fiom the frequency distribution of the sample mean presented earlier, is equal to 10. In Table 16.3, we have used the relative fi-equency for the calculation of the probabilities. 16.8 STANDARD ERROR OF A STATISTIC In the previous Section we learnt that we can draw a number of samples depending upon the population and sample sizes. From each sample we get a different value for the statistic we q e looking for. These values can be arranged in the form of a probability distribution, which is called the sampling distribution of the concerned ' statistic. The statistic is also similar to a random variable since a probability is attached to each value it takes. In Table 16.3 in the previous Section we have presented the statistic along with its probability. Ssnipling Theory and Survey Techniques We have learnt in Unit 14 that mathematical expectation of a random variable is equal to its arithmetic mean. Let us find out the mathematical expectation and standard deviation of the sampling distribution. We notice two important properties of the sampling distribution. 1) The expectation of the sampling distribution of the statistic is equal to the population parameter. Thus if we have the sampling distribution of sample means, then its expected value is equal to population mean. Symbolically, E ( T ) = P. 2) The standard deviation of the sampling distribution is called 'standard m r ' of the concerned statistic. Thus if we have sampling distribution of sample means, then its standard deviation is called the 'standard error of sample means'. Thus stand&mr indicates the spread of the sample means away fiom the population mean. In Block 7 we would see that standard error is used for hypothesis testing and statistical estimation. Example 16.3 Find out the standard error of the sampling distribution given in Table 16.3 We know that standard error of the sample mean is standard deviation of the sampling distribution. Thus, J-=--=T ox= E(x)' -['E(x)] Now, and Thus, the standard error of the sample mean in this case is 1.73. Now a question may be shaping up in your mind. Do we have to draw all possible samples to find out standard m r ? In Example 16.3 above we £kt noted down all the possible samples, m g e d these in a dative frequency distribution form and thereaRer calculated the standard deviation. In Example 16.3 the population size and sample size were quite small, and thus the task was manageable. But, can you imagine what would happen when we have much larger population and sample sizes? It is too difficult and cumbersome a task. In fact the entire advantages of sampling disappears if we start selecting all possible samples! Secondly, is it possible to fit a theoretical probability distribution (discussed in Block 5) to the sampling distribution? In fact, the Central Limit Theorem says that, "if samples of size n are drawn from any population, the sample medm are approximately normally distributed for large values of n". Thus whatever be the distribution of the population, the sampling distribution of p will approximately normal for large enough sample sizes. If the population is normad then sampling  distribution of x is normal for any sample size. If populatiorl is approximately Basic Concepts of Sampling normally distributed than sampling distribution of.T is approximately normal even for small sample size. Moreover, even if population is not normally distributed, sampling distribution of is approximately n o d a l for large sample sizes. Th~dly,what is the relationship between standard deviation of the population h m which the sample is drawn and the standard error of F ? Obviously, the spread of x will be less than the spread of the population units. The standard error of x is given by 0 0,= 6 , ' where al is standard error of p and a is standard deviation of the original population. Thus standard error is always smaller in value than standard deviation of the population, because standard error is equal to the standard deviation of the population divided by square root of the sample size. The above is true for simple random sampling with replacement. When sampling is without replacement in that case we have to make some finite population a N-n correction and standard error is given by OX=- x- J;; ?I-1. n When the ratio - is very small both the procedures give almost similar results. N But when sample size is not neghgible compared to population size the correction factor needs to be applied. How do we interpret the standard error? As mentioned earlier it shows the spread I of the statistic. Thus, if standard error is smaller then there is a greater probability that the estimate is closer to the concerned parameter. Example 16.4 , Consider the population: 2,5,8,13 i) Calculate the population mean and the population standard deviation. ii) Construct a sampling distribution of the sample mean when random samples j of size 2 are selected from the population a) with replacement, and I b) without replacement. Find the mean and the standard error of the distribution in each case. I iii) Verify that in the case of random sampling with replacement, E (5) = p and oz=- a and in the case of random sampling without r e i , h e n t , E(5) = p I & I Answer: We have the population: 2,5,8,13; population size N = 4; sample size n = 2. - - -. - -.- Sampling Theory and Survey Techniques Population standard deviation: ii) (a) Number of possible samples with replacement = N" = 16. The samples: (2,219 (2,513 (2,819 (2,131, (5,219 (5,513 (5,819 (5,131, (8921, (8,513 (8981, (8,131, (13921, (13,51, (13,819 (13,131. The sample means: 2, 3.5, 5, 7.5, 3.5, 5, 6.5, 9, 5, 6.5, 8, 10.5, 7.5, 9, 10.5, 13. Sarqiling.Distributionof Sample Means:  Basic Concepts of Mean of the sampling distribution: Snmpllng E(x) = zn i= l eFi (where Ti is the mean of ifh sample) Standard.error of the distribution: oi9=, / E ( Z ~-) { E ( Z ) } ~ Now, 1 =-%748=57.31. 16 And, { E ( z )=) 7'~ = 49 OF= 31-49=f i = 2.83. Thus, the mean and the standard error of the sampling distribution in the case of random sampling with replacement are 7 and 2.83 respectively. b) Number of possible samples without replacement = Cn =4 C2 = 6 The samples: (2Y5)Y (2Y8)Y (2Y13)Y (57817 (5713)Y (8913). Sample means: 3.5, 5, 7.5, 6.5, 9, 10.5. Sampling Distribution of Sample Means; \81:1!t1;11grtkeory and hurbc*} Techniques Mean of the population: Standard error of the distribution Now, And we already know, Thus, the mezn and the standard error of the sampling distribution in the case of random sampling without replacement are 7 and 2.35 respectively. iii) In the case of random sampling with replacement, E(Z)= 7 = p as we have independently obtained fiom the sampling distribution of the sample means. In the case of random sampling without replacement, Basic Concepts o t Sanipli~~g as we have independently obtained from the sampling distribution of the sample mean Hence, our results are all verified. 16.9 DESIRABLE PROPERTIES OF AN ESTIMATOR Suppose, 8 is an unknown population parameter that we are interested in. We may want to estimate 8on the basis of a random sample drawn kom the population. For this purpose we may use a statistic T (which is a h c t i o n of the sample values). Here T is an estimator of 8 and the value of T that is obtained from the given sample is an estimate of 8. In fact, the value is known as apoint esRmate in the sense that it is one particular value of the estimator (see Unit 18 for details). Earlier, we have discussed the concepts of sampling and non-sampling errors. We recapitulate here that the absolute difference (ignoring the sign) between a sample IT statistic and the population parameter, i.e., - 81 measures the extent of the sampling error. We may note here that an estimator is essentially a formula for computing an estimate of the population parameter and there can be several potential estimators (alternative formulae) that may be used for this purpose. So, there should be some desirable properties on the basis of which we can select a particular estimator for estimating the population parameter. A very simple requirement for IT 4 T to be a good estimator of 8 is that the difference - should be as small as possible. Various approaches have been suggested to ensure this. 16.9.1 Unbiasedness We have already noted that the value of a statistic varies from sample to sample due to sampling fluctuation. Although the individual values of a statistic may be different from the unknown population parameter, on an average, the value of a statistic should be equal to the population parameter. In other words, the sampling distribution of T should have a central tendency towards 8. This is known as the property of unbiasedness of an estimator. It means that although an individual value of a given estimator may be higher or lower than the unknown value of the population parameter, there is no bias on the part of the estimator to have values that are always greater or smaller than the unknown population parameter. If we accept that mean (here, expectation) is a proper measure for central tendency, then T is an unbiased estimator for 8 if E(T) = 8. 16.9.2 Minimum-Variance It is also desirable that the average spread of all the possible values of an unbiased. estimator around the population parameter is as small as possible. It will reduce the chance of an estimate being far away from the parameter. If we accept that variance is a proper measure for average spread (dispersion), we want that among all the unbiased estimators, T should have the smallest variance. Symbolically, V (T) 5 V (T' ) where, V stands for variance and T' is any other-unbiasedestimator. Santpling Thcory and An estimator T, which is unbiased and among all the unbiased estimator has the Survcy Techniques minimum variance, is known as a minimuy-variance unbiased estimator. Let us consider an example. Suppose, we have a random sample of size n from a given population of size N. In this case, the sample mean is given by * =; 1 " x x t Ci * where x, is the i* member of the sample. It can be proved that it is an unbiased estimator of the population mean p. Symbolically E(F) = p However, it carbe shown that the sample variance defined as s2= -c 1 " n ,=I -PY (3 is not an unbiased estimator of the population variance aZ. Symbolically, E(s2)f: a2 1 " On t h ~.I, ltrary, if we define the sample variance as s12= - n - 1 ,. (XI-rY,then sr2is an unbiased estimator of = - 1 N-1 f (x, - z ) ~ , l=l Suppose M e r that the sample values are not only random but also independent (random sample with replacement) and the underlying population is normal. It can be shown that the sample mean is not only an unbiased estimator of the population mean p but also it has the minimum variance among all the unbiased estimators of p. 16.9.3 Consistency and Efficiency Another approach may be to suggest that the estimator T should approximate the unknown population parameter 8 as the sample size n increases. Since T itself is a random variable, we may express this requ&ment in probabilistic or stochastic terms as the statistic T should converge to the parameter 8 stochastically (i.e., in probability) as n j lw. A statistic T with this property is called a consistent estimator of 8. In real life, a large number of consistent estimators of the same parameter q have often been found. In such a situation, obviously, some additional criterion is needed to choose among these consistent estimators.One such criterion may be to demand that not only T should converge stochastically to 8but also it should do so quite rapidly. Without going into the details, we may mention here that some times an estimator assumes the form of a normal distribution when the sample size n increases indefinitely. Such estimators are called usymptotically normal. If we focus on consistent estimators that are asymptotically n o d , the rapidity of their convergence is indicated by their respective asymptotic variances. In fact, the convergence is the fastest for the estimator that has the lowest asymptotic variance. Such kind of an estimator is known as an efficient estimator among all the asymptotically normal consistent estimators of a population parameter. Check Your Progress 2 1) Define the following concepts: a) Simple Random Sample  Basic concepts of b) SamplingDistribution Sampling c) Standard Error................................................................................................................... 2)' Distinguish between the folloking: a) Simple rkdom samplinggwithreplacement and Simple random sampling without replacement b) Simple random sampling and stratified random sampling................................................................................................................... 3) Given a population: 1,2,5,6. Bring out all possible'samples of size 2 i) with replacement, and ii) without replacement. 4) Given a population: 2,4,6. Suppose a sample of size 2 is to be selected from this population by the method of random sampling without replacement.. a) Present the sampling distribution of sample mean. b) Compute the standard error. 16.10 LET US SUM UP i In this unit, we distinguished between the census method and the sample method of conducting a itatistical inquiry. We have seen that on account of various r e s o w constraints, census method cannot be undertaken always. Moreover, due to the enormity of the task involved, the chances of committing non-sampling mrs in I a census are at times quite high. A sample survey, on the other hand, has some definite advantages. A properly conducted sample survey is generally less error prone. A sample survey has a sound scientific basis. ~ s ' result, a the sampling distribution of @e relevant statistic (obtained from random samples) forms an objective basis of assessment about a population parameter. Sampling Theory and Survey Techniques 16.11 KEY WORDS Estimate : It is the articular value that can be obtained fiom an estimator. Estimator : It is the specific functional fonn of a statistic or the formula involved in its cal&cm, Gendy, the two tenns, statistic and estimator, are used interchangeably. Parameter : It is a measure of some characteristic of the population. Population : It is the entire collection of units of a specified type in a given place and at a particular point of time. Random Sampling : It is a procedure where every member of the population has a definite chance or probability' of being selected in the sample. It is also called probability sampling. Sample : It is a sub-set of the population. Therefore, it is a collection of some units h m the population. Sampling Distribution : It refers to the probability distribution of a statistic. Sampling Error : The absolute difference between population parameter and relevant sample statistic. Sampling Fluctuation : It is the variation in the values of a statistic computed fiom different samples. Simple Random Sampling : This is a sampling procedure, in which, each member of the population has the same chance of being selected in the sample. Standard Error : It is the standard deviation of the sampling distribution of a statistic. Statistic : It is a function of the values of the units that are included in the sample. The basic purpose of a statistic is to estimate some population parameter. Statistical Inference : It is the process of drawing conclusions about an unknown population characteristic on the basis of a known sample drawn from it. 16.12 SOME USEFUL BOOKS Bhardwaj. R. S., 1999, Business Statistics (First ~dition),Excel Books, New Delhi, Chapter 20. Nagar, A, L. and Dar, R. K., 19E8,Basic Statistics, Oxford University Press, Delhi, Chapter 9. Goon, -4.M., Gupta, 34.K.and Dasgupta, B., 1971, Fz4ndantentczls qfStntistics. Volume 1 (Fourth Edition), The World Press Private Limited, Calcutta, Chapters Basic Concepts of Sampling 14, 15 and 16. 16.13 ANSWERS/HINTS TO CHECK YOUR PROGRESS EXERCISES Check Your Progress 1 1) Read the text and define these terms in one or two sentences each. 2) Read the text and distinguish in a few sentences,. 3) Read'the text and answer in a few sentences. Check Your Progress 2 1) Read the text and define these terns in a few sentences. 2) Read the text and distinguish in a few sentences. 3) Go through Example 16.2 in the text and attempt yourself. 4) 0.82.