Generalized modified linear systematic sampling scheme for finite populations

Hacettepe Journal of Mathematics and Statistics Volume 43 (3) (204), 529 542 Generalized modified linear systematic sampling scheme for finite populations J Subramani and Sat N Gupta Abstract The present paper deals with a further modification on the selection of linear systematic sample, which leads to the introduction of a more generalized form of modified linear systematic sampling namely generalized modified linear systematic sampling (GMLSS) scheme, which is applicable for any sample size, irrespective of the population size whether it is a multiple of sample size or not The performances of the proposed modified linear systematic sampling scheme are assessed with that of simple random sampling, circular systematic sampling for certain hypothetical populations as well as for some natural populations As a result, it is observed that the proposed modified linear systematic sample means perform better than the simple random sample mean and circular systematic sample mean for estimating the population mean in the presence of linear trend among the population values Further improvements on GMLSS are achieved by introducing Yates type end corrections Keywords: Linear Trend, Modified Systematic Sampling, Natural Population, Simple Random Sampling, Trend Free Sampling, Yates End Corrections, Yates Type End Corrections 2000 AMS Classification: 62D05 Introduction It is well known that in the presence of linear trend among the population values the systematic sampling performs better than the simple random sampling without replacement for estimating the population mean Later several modifications have been made to improve the efficiency of the systematic sampling by introducing changes on the method of selection which includes centered systematic sampling by Madow [2], balanced systematic sampling by Sethi [5], modified systematic Department of Statistics, Pondicherry University, R V Nagar, Kalapet, Puducherry 60504, India, Email: drjsubramani@yahoocoin Corresponding Author Department of Mathematics and Statistics, The University of North Carolina at Greensboro, Greensboro, NC 27402, USA, Email: sngupta@uncgedu

sampling by Singh, Jindal, and Garg [6] and changes on the estimators itself like Yates end corrections [26] In all these sampling schemes mentioned above and also in the case of linear systematic sampling (LSS), it is assumed that the population size N is a multiple of sample size n and there is no restriction on the part of sample size For a detailed discussion on estimation of finite population means, one may refer to Bellhouse and Rao [2], Chang and Huang [3], Cochran [4], Fountain and Pathak [5], Gupta and Kabe [6], Kadilar and Cingi [8], Khan et al [9] [], Murthy [4], Singh S [7], Subramani [8] [9] [20] [2], Subramani and Singh [24] and the references cited therein The circular systematic sampling (CSS) is an alternative to LSS whenever the population size is not a multiple of sample size with certain restrictions to get distinct units in the sample For a discussion on the choices for the sampling interval for the case of CSS one may refer to Bellhouse [], Sudakar [25] and Khan et al [0] Chang and Huang [3] have suggested a modification on linear systematic sampling and introduced the Remainder Linear systematic sampling (RLSS) The RLSS can be used for population size is not a multiple of sample size, where N = n k + r, k is the sampling interval for RLSS and depends upon the remainder r When the remainder is zero, the RLSS reduces to the usual linear systematic sampling Recently Subramani [22] [23] has introduced a modification on the selection of a systematic sample in the linear systematic sampling by choosing two random starts and is called as modified linear systematic sampling (MLSS) scheme However the problem is that the MLSS is also applicable only for the cases where the population size is a multiple of sample size and is not valid when the population size is not a multiple of sample size In this paper, a more general form of the MLSS is introduced which is applicable for any sample size whether it is even or odd and for any population size N, where N = nk or N nk, where k is the sampling interval For the sake of convenience, it is assumed that, without loss of generality, that N = n k + n 2 k 2 such that n = n + n 2 and the value of n n 2 Further, it is shown that the MLSS method discussed by Subramani [22] [23] and the usual LSS are the particular cases of the proposed methods of generalized modified systematic sampling The explicit expressions for the GMLSS sample means, the bias and the mean squared error are obtained for certain hypothetical populations with a perfect linear trend among the population values and are compared with that of simple random sampling Further the relative performances of GMLSS are assessed with that of the simple random sampling, circular systematic sampling and remainder linear systematic sampling for certain natural populations From the numerical comparison, it has been shown that the GMLSS perform better than the simple random sampling and circular systematic sampling for estimating the finite population means in the presence of linear trend The entire above are explained with the help of hypothetical as well as natural populations 2 Proposed Modified Systematic Sampling Scheme for any sample size The proposed modified systematic sampling scheme is explained here, firstly with the help of examples and later the generalized case For the sake of simplicity and for the benefit of the readers, selecting a modified linear systematic sample of

size 5 from a population of size N = 2 and N = 6 is explained with the help of following examples 2 Example Let N = 2, n = 5, n = 3, k = 2, n 2 = 2, k 2 = 3 Step : Arrange the 2 population units as given below: Arrangement of the Population units i j 2 2 3 2 3 4 5 6 7 8 9 0 2 Step 2: Select two random numbers i 2 and j 3; include all the elements in the columns corresponding to i and j The selected samples are given below: GMLSS Samples Sample No i j Sampled Units,3,6,8, 2 2,4,6,9, 3 3,5,6,0, 4 2 2,3,7,8,2 5 2 2 2,4,7,9,2 6 2 3 2,5,7,0,2 22 Example Let N = 2, n = 5, n = 3, k = 4, n 2 = 2, k 2 = 2 Step : Arrange the 6 population units as given below: Arrangement of the Population units i j 2 3 4 2 2 3 4 5 6 7 8 9 0 2 3 4 5 6 Step 2: Select two random numbers i 4 and j 2; include all the elements in the columns corresponding to i and j The selected samples are given below:

GMLSS Samples Sample No i j Sampled Units,5,7,,3 2 2,6,7,2,3 3 2 2,5,8,,4 4 2 2 2,6,8,2,4 5 3 3,5,9,,5 6 3 2 3,6,9,2,5 7 4 4,5,0,,6 8 4 2 4,6,0,2,6 2 Generalized Modified Linear Systematic Sampling Scheme (GMLSS) The steps involved in selecting a generalized modified linear systematic sample (GMLSS) of size n from a population of size N = n k + n 2 k 2, where n = n + n 2, k and k 2 such that k = k + k 2 are positive integers, are as follows: Firstly arrange the N population units (labels) in a matrix with k = k + k 2 columns as given in the Fig That is, the first n 2 k population units are arranged row wise in the first n 2 rows with k elements each and the remaining (n n 2 )k population units are arranged row wise in the next (n n 2 ) rows with k elements each as in the arrangement given below in the Fig 2 The first k columns are assumed as Set and the next k 2 columns are assumed as Set 2 3 Select two random numbers, i in between and k and j in between and k 2, then select all the n units in the i th column of Set and all the n 2 units in the j th column of Set 2, which together give the sample of size n 4 The step 3 leads to k k 2 samples of size n Figure Arrangement of the population units i k k + k + j k k + k + i k + k k + k + k + k + j 2k 2k + 2k + i 2k + k 2k + k + 2k + k + j 3k (n 2 )k + (n 2 )k + i (n 2 )k + k (n 2 )k + k + (n 2 )k + k + j n 2 k n 2 k + n 2 k + i n 2 k + k n 2 k + k + n 2 k + k + i n 2 k + 2k n 2 k 2 + (n )k + n 2 k 2 + (n )k + i n 2 k 2 + n k Since the generalized modified linear systematic sampling scheme has k k 2 samples of size n and each unit in the Set- is included in k samples and each unit in the Set-2 is included in k 2 samples, the first order and second order inclusion probabilities are obtained as given below:

(2) π i = { k k 2 if i th unit is from the Set-, if i th unit is from the Set-2 k if i th and j th units are from the same column of the Set-, (22) π ij = k 2 if i th and j th units are from the same column of the Set-2, k k 2 if i th and j th units are from Set- and Set-2 respectively, 0 Otherwise In general, for the given population size N = n k + n 2 k 2, the selected generalized modified linear systematic samples (labels of the population units) for the random starts i and j are given below: (23) S ij = { i, i + k, i + 2k,, i + (n 2 )k, i + n 2 k, i + n 2 k + k,, i + n 2 k + (n n 2 )k, j + k, j + k + k,, j + k + (n 2 )k (i =, 2, 3,, k and j =, 2, 3,, k 2 ) The generalized modified linear systematic sample mean based on the random starts i and j is obtained as (24) ȳ gmlss = ȳ ij = n ( n2 l=0 n n 2 y i+kl + l=0 n 2 y i+n2k+k l + l=0 (i =, 2, 3,, k and j =, 2, 3,, k 2 ) y j+k+kl Since the first order inclusion probabilities are not equal, the generalized modified linear systematic sample mean given above in Equation (24) is not an unbiased estimator The mean squared error of the GMLSS mean can be obtained from the Equation (25) given below: (25) MSE (ȳ gmlss ) = k k 2 k k 2 (ȳij Ȳ ) 2 i= j= 3 Population with a Linear Trend As stated earlier the linear systematic sampling (LSS) has less variance than the simple random sampling if the population consists solely of a linear trend among the population values In this section, the relative efficiencies of the generalized modified systematic sampling schemes with that of simple random sampling for estimating the mean of finite populations with linear trend among the population values are assessed for certain hypothetical populations In this hypothetical population, the values of N = n k +n 2 k 2 population units are in arithmetic progression That is, ) (3) Y i = a + ib (i =, 2, 3,, N)

After a little algebra, one may obtain the generalized modified systematic sample means (GMLSS) with the random starts i and j and the population mean for the above hypothetical population are as given below: (32) ( n i + n 2 j + n 2 k + (k + k 2 )n 2 (n ) + k (n n 2 )(n n 2 ) 2 ȳ gmlss = ȳ ij =a + n (i =, 2, 3,, k and j =, 2, 3,, k 2 ) ) b ( ) k n + k 2 n 2 + (33) Ȳ = a + b 2 For the above population with a linear trend, variances of the simple random sample mean V (ȳ r ) together with the Bias and Mean Squared Error of the generalized modified linear systematic sample means Bias (ȳ gmlss ) and MSE (ȳ gmlss ) are obtained as given below: (N n)(n + )b2 (34) V (ȳ r ) = 2n Bias (ȳ gmlss ) = (k k 2 )n 2 {n (n 2 + )} (35) b 2n MSE (ȳ gmlss ) = { (36) 2n 2 n 2 (k 2 ) n 2 2(k2 2 ) + 3n 2 2 {n (n 2 + )} 2 (k k 2 ) 2} b 2 Since the algebraic comparisons of the various expressions of the variance and the mean squared error given in Equations (34) and (36) are not possible due to the presence of several different parameters, we have compared them numerically and are presented in the Table 3 3 Remark If we put k = k 2 in Equation (35), then Bias (ȳ gmlss ) = 0 That is, the Generalized modified linear systematic sample means become the modified linear systematic sample means of Subramani (203a, b) and the resulting estimators are unbiased 32 Remark Even when k k 2 if we put n = n 2 + in Equation (35), the Bias (ȳ gmlss ) = 0 That is, the GMLSS estimators are unbiased estimators of the population mean 33 Remark When n 2 = 0 in Equation (35), the Bias (ȳ gmlss ) = 0 When n 2 = 0 which means n = n, Equation (36) reduces to the variance of linear systematic sample mean That is, the GMLSS estimators are unbiased estimators of the population mean and the resulting generalized modified linear systematic sample means become the linear systematic sample means 34 Remark The mean squared error of the Remainder Linear Systematic sample mean is given below: (37) MSE (ȳ rlss ) = k( k + ) k k+ (ȳij Ȳ ) 2 i= j= where k is the sampling inverval for RLSS

35 Remark Under the Remainder Linear systematic sampling procedure, the variance of Horvitz-Thompson estimator obtained by Chang and Huang [3] is given below: (38) V rlss (ȳ HT ) = N 2 k (n r)2 k2 k i= where ȳ i is the i th sample mean of the Set- V gmlss (ȳ HT ) = N 2 ȳ 2i is the i th sample mean of the Set-2 ) 2 (ȳi Ȳ + r2 ( k + ) 2 Ȳ is the population mean of the first stratum Ȳ 2 is the population mean of the second stratum k + k+ i= ) 2 (ȳ2i Ȳ2 36 Remark Under the Generalized modified linear systematic sampling, the variance of Horvitz-Thompson estimator is given below: k k 2 (39) n2 k2 k i= ) 2 (ȳi Ȳ + n2 2 k2 2 where ȳ i is the i th sample mean of the Set- ȳ 2i is the i th sample mean of the Set-2 Ȳ is the population mean of the Set- Ȳ 2 is the population mean of the Set-2 k 2 i= ) 2 (ȳ2i Ȳ2 37 Remark When n = (n r), k = k, n 2 = r, and k 2 = k +, the generalized modified linear systematic sampling reduces to Remainder Linear systematic sampling for the population with a perfect linear trend This shows that RLSS is the particular case of GMLSS for the population with a perfect linear trend However the arrangement of the population units are different and hence the Set (2) and Stratum (2) are not one and the same 4 Some Modifications on Generalized Modified Linear Systematic Sample Means It has been shown in Section 3 that the estimators based on the generalized modified linear systematic sampling schemes (GMLSS) are, in general, not unbiased estimators However, further improvements can be achieved by modifying the generalized modified linear systematic sample means as done by Subramani [2] in the case of modified linear systematic sample mean by introducing Yates type end corrections [26] Consequently the proposed sampling scheme becomes completely trend free sampling (Mukerjee and Sengupta [3]) 4 Yates Type End Corrections for GMLSS Means The modification involves the usual generalized modified linear systematic sampling, but the modified sample mean is defined as (4) ȳ gmlss = ȳ gmlss + β(y y n ) = Ȳ That is, the units selected first and last are given the weights n + β and n β respectively, whereas the remaining units will get the equal weight of n, so as to

make the proposed estimator is equal to the population mean That is, the value of β is obtained as (42) β = Ȳ ȳ gmlss (y y n ) For the hypothetical population defined in Section 3, after a little algebra, we have obtained the value of β for the two random starts i and j as given below: (43) β = (n n 2 ) {k (n 2 + ) k 2 n 2 } + 2 {k 2 n 2 n (i + ) n 2 (j + )} 2n {(i j) (k + k(n 2 ))} In the similar manner one can propose Yates type end corrections, alternate to Yates end corrections by giving different weights to two different units other than the first and last units For example one may give different weights to two successive units; the first units of the two subsamples and so on If we give different weights to the two successive units between and n 2 in Set or successive units in Set 2 then the revised estimator ȳgmlss for the case of generalized modified linear systematic sample mean ȳgmlss = ȳ gmlss + β (y l y l+ ) = Ȳ, which yields the value of β as is given below: (44) β = Ȳ ȳ gmlss (y l y l+ ) = (n 2 n ) {k (n 2 + ) k 2 n 2 } 2 {k 2 n 2 n (i + ) n 2 (j + )} 2nk If we give different weights to the two successive units between n 2 + and n in Set or the first units of the two sets then the revised estimator ȳgmlss for the case of generalized modified linear systematic sample mean ȳgmlss = ȳ gmlss + β 2 (y l y l +) = Ȳ, which yields the value of β 2 as is given below: (45) β 2 = Ȳ ȳ gmlss (y l y l +) = (n 2 n ) {k (n 2 + ) k 2 n 2 } 2 {k 2 n 2 n (i + ) n 2 (j + )} 2nk If we give different weights to the first and the last units in the Set then the revised estimator ȳgmlss for the case of generalized modified linear systematic sample mean ȳgmlss = ȳ gmlss + β 3 (y y n ) = Ȳ, which yields the value of β 3 as is given below (46) β 3 = Ȳ ȳ gmlss (y y n ) = (n 2 n ) {k (n 2 + ) k 2 n 2 } 2 {k 2 n 2 n (i + ) n 2 (j + )} 2n {k 2 n 2 + (n )k } If we give different weights to the units n + and n in Set 2 (ie the first and the last units in Set 2), then the revised estimator ȳgmlss for the case of generalized modified linear systematic sample mean ȳgmlss = ȳ gmlss + β 4 (y l y l +) = Ȳ, which yields the value of β 4 as is given below: (47) β 4 = Ȳ ȳ gmlss (y y n2 ) = 2(n i + n 2 j) N + n + n n 2 (k 2 k ) 2nk(n 2 )

4 Remark In the presence of a perfect linear trend, the revised generalized modified linear systematic sampling estimators ȳgmlss = ȳ gmlss = ȳ gmlss = ȳgmlss = ȳ gmlss = Ȳ and hence the variances are zero That is, GMLSS becomes a trend free sampling (Mukerjee and Sengupta [3]) 5 Relative Performance of Modified Linear Systematic Sampling for Certain Natural Populations The Generalized proposed modified linear systematic sampling schemes were introduced in Section 2, where as its means, bias and the mean squared error were derived in Section 3 for the hypothetical populations with a perfect linear trend among the population values Further it has been shown (Remarks 3 to 33) that the usual linear systematic sampling (LSS) and the modified linear systematic sampling (MLSS) schemes are the particular cases of the proposed GMLSS scheme Further it has been shown (Remark 37) that the RLSS is the particular case of the GMLSS for the population with a perfect linear trend Since the expression of the mean squared error of GMLSS means is involved several parameters compared to simple random sampling without replacement, it is not feasible to make an algebraic comparison Hence we have assessed the performances of GMLSS means with that of SRSWOR means, CSS means and RLSS means for a hypothetical population together with some natural populations considered by Murthy, p228 [4] The data were collected for estimating the output of 80 factories in a region The data pertaining to the number of workers, fixed capital and the output are respectively denoted as population 2, population 3 and population 4, where as its labels (presuming a hypothetical population with a perfect linear trend, as considered as population It is already established in Subramani [22] [23] that whenever the population size is a multiple of sample size the modified linear systematic sampling performs well compared to simple random sampling as well as linear systematic sampling Hence we have obtained the variance of the simple random sample mean V (ȳ r ), the variance of the circular systematic sample mean V (ȳ css ), the mean squared error of the remainder linear systematic sample mean, MSE (ȳ rlss ), the mean squared error of the generalized modified systematic sample mean, MSE (ȳ gmlss ), the variance of Horvitz Thompson estimator under RLSS, V rlss (ȳ HT ) and the variance of Horvitz Thompson estimator under GMLSS, for the odd sample sizes from 7 to 25 so as the population size 80 is not a multiple of the sample size with various possible combinations of n and n 2 such that n > n 2 and are presented in Table 3 From the table values, it is seen that, out of the 40 cases considered the generalized modified linear systematic sample (GMLSS) means perform better than simple random sample mean and circular systematic sample mean in all the 40 cases Further it is observed from Equation (36) that whenever the differences between k and k 2 and n and n 2 approaches zero simultaneously, then the efficiency of the GMLSS improves It is to be noted that for the fixed population size N = 80 there are 0 different choices for the sample of size n = The mean squared error of the generalized modified linear systematic sample mean under the possible combinations of k, n, k 2 and n 2 for population (population with a perfect linear trend) is given in the following Table :

Table MSE ( ȳ gmlss ) under the possible combinations of k, n, k 2 and n 2 for population N n k n k 2 n 2 MSE (ȳ gmlss ) 7 0 0 456 4 9 22 2 9856 6 9 3 2 699 8 9 4 2 832 80 4 8 6 3 4508 7 8 8 3 280 4 7 3 4 307 8 7 6 4 304 0 6 4 5 27 5 6 0 5 230 From the possible combinations of k, n, k 2 and n 2, one can choose the optimum value of k, n, k 2 and n 2 such that the differences between k and k 2, and n and n 2 approaches zero simultaneously The different sample sizes with the population size N = 80 together with the sampling interval of Circular systematic sampling, Remainder linear systematic sampling and Generalized modified linear systematic sampling considered for the numerical comparisons of these sampling schemes are given below: Table 2 Sample size and sampling interval of Circular systematic sampling (CSS), Remainder linear systematic sampling (RLSS) and Generalized modified linear systematic sampling (GMLSS) N n CSS RLSS GMLSS k ˆk n r r ˆk + k n k 2 n 2 7 4 3 2 4 2 3 9 9 8 8 9 8 5 0 4 7 7 8 3 8 5 6 0 5 3 6 6 2 7 8 7 4 6 80 5-5 0 5 6 3 8 8 7 7 5 4 5 2 5 4 6 6 9 4 4 5 4 5 4 5 5 4 2-3 4 7 4 4 7 3 4 23 3 3 2 4 3 2 4 25 3 3 20 5 4 3 20 4 5 where k is Sampling interval for Circular Systematic Sampling and k is Sampling interval for Remainder Linear Systematic Sampling

Table 3 Comparison of simple random sample, Circular systematic sample, Remainder Linear systematic sample, Horvitz-Thompson estimator based on RLSS, Generalized modified linear systematic sample means and Horvitz- Thompson estimator based on GMLSS for the four natural populations considered by Murthy (967), page 228 The four populations are Population -Labels (presuming a hypothetical population with a perfect linear trend), Population 2-Number of workers, Population 3-Fixed capital and Population 4-The output Population Number 2 3 4 n V (ȳ r) V (ȳ css) MSE(ȳ rlss ) V rlss (ȳ HT ) MSE(ȳ gmlss ) V gmlss (ȳ HT ) 7 7039 53 545 544 545 544 9 5325 659 533 545 325 338 4234 507 25 243 230 350 3 3479 34 28 2 79 269 5 2925 2 9 33 9 7 250 384 0 20 089 097 9 267 273 087 083 087 083 2 896 084 092 084 092 23 673 099 047 05 047 05 25 485 285 048 045 048 045 7 953359 90846 49064 63859 07556 02433 9 7286 2789 08976 75 5705 55567 57344 85699 40387 47099 40380 56945 3 4754 58535 28444 32285 337 543 5 39644 28234 35248 2293 25362 7 338784 60559 29484 33273 955 5840 9 293500 42832 3684 747 2343 2099 2 25684 20099 2255 9686 2694 23 226558 45865 502 979 9533 950 25 2020 42057 7582 059 3826 273 7 932263 924493 237970 34846 07063 02674 9 705245 6480 9962 04790 57693 580958 5606687 83936 28466 307449 672462 849278 3 4606609 724769 3428 328075 422709 65968 5 387329 3759 45523 247996 372577 7 33239 845746 43728 49098 56898 5296 9 2869632 402956 237367 286438 93437 75956 2 25208 3368 47022 53475 68277 23 2258 527704 82073 07023 79758 73963 25 966403 496774 32632 464 278579 24595 7 43925684 257379 790222 779205 802238 757872 9 33228409 6780868 3788733 375529 3290760 3007034 2642053 5356957 2068943 2022072 2547605 286335 3 2708267 3704846 647984 635528 2457986 3984635 5 8252225 27087 3380 30236 22442 7 5609369 3348979 03502 062527 862255 98240 9 3522903 2320385 037255 035309 46890 344696 2 833860 588670 63200 02766 20630 23 0438563 6403667 363285 42970 632486 63369 25 926654 936725 250705 266348 859895 76407

Table 4 Efficiency of the Generalized modified linear systematic sample means and Horvitz-Thompson estimator based on GMLSS for the 4 natural populations Population Number 2 3 4 n R R 2 R 3 R 4 R 5 R 6 R 7 R 8 R 9 7 292 22 00 00 294 22 00 00 00 9 638 203 64 68 575 95 58 6 04 84 220 09 06 20 45 072 069 52 3 944 9 22 8 293 27 08 078 50 5 299 09 089 2458 02 00 089 7 280 43 24 35 2578 396 3 24 09 9 249 34 00 095 26 329 05 00 095 2 2257 00 0 206 09 00 0 23 3560 2338 00 09 3280 255 092 00 09 25 3094 594 00 094 3300 633 07 00 094 7 886 77 39 52 93 86 46 60 095 9 264 98 9 96 298 203 96 20 097 420 22 00 7 007 50 07 083 4 3 420 76 086 097 92 4 056 063 54 5 728 23 54 562 39 7 769 36 54 74 239 382 86 20 083 9 268 85 059 075 403 205 065 083 090 2 305 02 3 84 093 02 0 23 2377 530 57 208 2385 535 58 208 00 25 455 304 055 076 652 345 062 086 088 7 87 80 6 26 908 87 2 3 096 9 233 204 73 78 24 200 7 75 02 834 2 042 046 660 096 033 036 26 3 090 7 074 078 698 0 048 050 56 5 562 52 84 040 0 22 50 7 2 539 279 32 266 553 286 32 097 9 483 208 23 48 63 229 35 63 09 2 636 087 096 492 079 087 0 23 2777 95 03 34 2995 2065 45 093 25 706 78 02 05 800 202 03 07 088 7 548 57 090 089 584 67 096 095 094 9 00 206 5 4 05 226 26 25 09 037 20 08 079 923 87 072 07 2 3 883 5 067 067 545 093 04 04 62 5 402 098 0 84 057 059 72 7 80 388 20 23 589 34 05 08 4 9 954 64 073 073 006 73 077 077 095 2 58 058 062 056 053 056 0 23 650 02 057 067 647 0 057 067 00 25 078 225 029 03 23 253 033 035 089 V (ȳ where R = r) MSE(ȳ gmlss ), R V (ȳ 2 = css) MSE(ȳ gmlss ), R 3 = V (ȳ R 5 = r) V gmlss (ȳ HT ), R 6 = and R 9 = V gmlss(ȳ HT ) MSE(ȳ gmlss ) MSE(ȳ rlss) MSE(ȳ gmlss ), R 4 = V rlss(ȳ HT ) MSE(ȳ gmlss ), V (ȳcss) V gmlss (ȳ HT ), R 7 = MSE(ȳ rlss) V gmlss (ȳ HT ), R 8 = V rlss(ȳ HT ) V gmlss (ȳ HT ),

6 Conclusion In this paper a generalized version of modified linear systematic sampling (GMLSS) scheme is introduced irrespective of the sample size, whether it is odd or even and the population size is not a multiple of the sample size The explicit expressions for the sample means and the mean squared error of the GMLSS estimators are derived for certain hypothetical populations with a perfect linear trend among the population values The performances of the proposed sampling scheme are assessed with that of the simple random sampling without replacement, Circular systematic sampling and Remainder linear systematic sampling for a hypothetical population together with some natural populations considered by Murthy [4] The comparative studies reveal that whenever there is a linear trend among the population values the generalized modified linear systematic sampling (GMLSS) performs well compared to simple random sampling In the case of circular systematic sampling, the sample size and the population size must be a co-prime to get all distinct units in the sample Hence the case with sample size 5 is not considered Further it is observed that when N = 80 and n = 2 with k = 4 the first unit and the last unit are the same and hence this case is also not considered From the above points, it is concluded that the proposed GMLSS is applicable for all possible combinations of population size and sample size even where the CSS fails In comparing with the RLSS, the GMLSS performs better than the RLSS in majority of the cases Acknowledgements The authors are thankful to the editor and the reviewers for their constructive comments, which have improved the presentation of the paper The first author wishes to record his gratitude and thanks to University Grants Commission for the financial assistance through UGC MRP References [] Bellhouse, D R On the choice of sampling interval in circular systematic sampling, Sankhya B, 247-248, 984 [2] Bellhouse, D R and Rao, J N K Systematic sampling in the presence of linear trends, Biometrika 62, 694-697, 975 [3] Chang, H J and Huang, K C Remainder linear systematic sampling, Sankhya B 62, 249-256, 2000 [4] Cochran, W G Sampling Techniques (3rd ed), New York: John Wiley and Sons, 977 [5] Fountain, R L and Pathak, P L Systematic and non-random sampling in the presence of linear trends, Communications in Statistics-Theory and Methods 8, 25-2526, 989 [6] Gupta, A K and Kabe, D G Theory of Sample Surveys, World Scientific Publishing Company, 20 [7] Horvitz, D G and Thompson, D J A generalization of sampling without replacement from a finite universe, Journal of the American Statistical Association 47, 663-685, 952 [8] Kadilar, C and Cingi, H Advances in Sampling Theory-Ratio Method of Estimation, Bentham Science Publishers, 2009 [9] Khan, Z, Gupta, S N and Shabbir, J A new sampling design for systematic sampling, Communications in Statistics-Theory and Methods 42 (8), 2659-2670, 203 [0] Khan, Z, Gupta, S N and Shabbir, J A Note on Diagonal circular systematic sampling, Forthcoming in Journal of Statistical Theory and Practice DOI: 0080/559860820380805, 203 [] Khan, Z, Gupta, S N and Shabbir, J Generalized Systematic Sampling, Forthcoming in Communications in Statistics-Simulation and Computation, 203

[2] Madow, W G On the theory of systematic sampling III-comparison of centered and random start systematic sampling, The Annals of Mathematical Statistics 24, 0-06, 953 [3] Mukerjee, R and Sengupta, S Optimal estimation of a finite population means in the presence of linear trend, Biometrika 77, 625-630, 990 [4] Murthy, M N Sampling Theory and Methods, Calcutta, India: Statistical Publishing House, 967 [5] Sethi, V K On optimum pairing of units, Sankhya B 27, 35-320, 965 [6] Singh, D, Jindal, K K and Garg, J N On modified systematic sampling, Biometrika 55, 54-546, 968 [7] Singh, S Advanced Sampling Theory with Applications, Springer, 2003 [8] Subramani, J Diagonal systematic sampling scheme for finite populations, Journal of the Indian Society of Agricultural Statistics 52 (2), 87-95, 2000 [9] Subramani, J Further results on diagonal systematic sampling for finite populations, Journal of the Indian Society of Agricultural Statistics 63 (3), 277-282, 2009 [20] Subramani, J Generalization of diagonal systematic sampling scheme for finite populations, Model Assisted Statistics and Applications 5, 7-28, 200 [2] Subramani, J A modification on linear systematic sampling for odd sample size, Bonfring International Journal of Data Mining 2 (2), 32-36, 202 [22] Subramani, J A modification on linear systematic sampling, Model Assisted Statistics and Applications 8 (3), 25-227, 203 [23] Subramani, J A further modification on linear systematic sampling for finite populations, Journal of Statistical Theory and Practice 7 (3), 47-479, 203 [24] Subramani, J and Singh, S Estimation of Population Mean in the Presence of Linear Trend, Forthcoming in Communications in Statistics-Theory and Methods, 203 [25] Sudakar, K A note on circular systematic sampling, Sankhya C 40, 72-73, 978 [26] Yates, F Systematic Sampling, Philosophical Transactions of the Royal Society A 24, 345-377, 948