A SPATIAL UNIT LEVEL MODEL FOR SMALL AREA ESTIMATION


 Joseph Miller
 2 years ago
 Views:
Transcription
1 REVSTAT Statistical Journal Volume 9, Number 2, June 2011, A SPATIAL UNIT LEVEL MODEL FOR SMALL AREA ESTIMATION Authors: Pero S. Coelho ISEGI Universiae Nova e Lisboa, Portugal Faculty of Economics, Ljubljana University, Slovenia Luís N. Pereira Escola Superior e Gestão, Hotelaria e Turismo, Centro e Investigação sobre o Espaço e as Organizações, Universiae o Algarve, Portugal Receive: May 2010 Revise: April 2011 Accepte: May 2011 Abstract: This paper approaches the problem of small area estimation in the framework of spatially correlate ata. We propose a class of estimators allowing the integration of sample information of a spatial nature. Those estimators are base on linear moels with spatially correlate small area effects where the neighbourhoo structure is a function of the istance between small areas. Within a Monte Carlo simulation stuy we analyze the merits of the propose estimators in comparison to several traitional estimators. We conclue that the propose estimators can compete in precision with competitive estimators, while allowing significant reuctions in bias. Their merits are particularly conspicuous when analyzing their conitional properties. KeyWors: combine estimator; empirical best linear unbiase preiction; small area estimation; spatial moels; unit level moels. AMS Subject Classification: 62D05, 62F40, 62J05.
2 156 P.S. Coelho an L.N. Pereira
3 A Spatial Unit Level Moel for Small Area Estimation INTRODUCTION Sample survey ata are extensively use to provie reliable irect estimates of parameters of interest for the whole population an for omains of ifferent kins an sizes. When the omains were not originally planne, they usually are poorly represente in the sample or even not represent at all. These omains are calle small areas an they usually correspon to small geographical areas, such as a municipality or a census ivision, or a small subpopulation like a particular economic activity or a subgroup of people obtaine by crossclassification of emographic characteristics. Traitionally, sample sizes are chosen to provie reliable estimates for large omains an the lack of sample ata from the target small area seriously affects the precision of estimates obtaine from areaspecific irect estimators. This fact has given rise to the evelopment of various types of estimators that combine both the survey ata for the target small areas an auxiliary information from sources outsie the survey, often relate to recent censuses an current aministrative ata, in orer to increase precision. Uner this context, the use of inirect estimators has been extensively applie. Such inirect estimators are base on either implicit or explicit moels that provie a link to relate small areas through auxiliary ata. Although traitional inirect estimators base on implicit moels, which inclue synthetic an composite estimators, are easy to apply, they usually present unesirable properties. For that reason, other moel base methos of small area estimation have been suggeste in the literature. These methos can make specific allowance for local variation through complex error structures in the moels that link the small areas, can be valiate from the sample ata an can hanle complex cases such as crosssectional, time series an spatial ata. Such methos are often base on explicit Linear Mixe Moels. The Best Linear Unbiase Preiction (BLUP) approach, using Henerson s metho ([13]), is the most popular technique for estimating small area parameters of interest (usually the mean or the total). Uner this approach an from the moel point of view the small area parameters of interest are functions of fixe (β) an ranom (u) effects. Consequently, the preiction of small area parameters of interest is base on the estimation/preiction of these moel effects. In practice this type of moels always involves unknown variance components in the variancecovariance structure of ranom effects. When these unknown components are substitute by consistent estimates the resulting estimator is usually name as Empirical Best Linear Unbiase Preictor (EBLUP). In the context of unit level spatial ata, little work has been one on moelbase methos of small area estimation. [25], [26] an [5] propose a spatial unit level ranom effects moel with spatial epenence incorporate in the error structure through a simultaneous autoregressive (SAR) error process.
4 158 P.S. Coelho an L.N. Pereira Other finings are ue to [4], [36], [30], [31], [32] an [40]. All these approaches consier a contiguity matrix to escribe the neighbourhoo structure between small areas. Nevertheless, there has been a lack of work regaring the explicit moeling of spatial correlation as a function of the istance between observations or small areas. The main aim of this paper is to propose an approach to the problem of small area estimation in circumstances in which the sample ata are of a spatial nature (or in other contexts in which it is possible to establish some kin of proximity between the omains of stuy), using an estimator that explicitly consier spatial correlation as a function of istance between small areas of stuy. This estimator, applicable to unit level ata, exploits both auxiliary information relating to other known variables on the population an structures of spatial correlation between the sample ata through the specification of an aequate noniagonal structure for the variancecovariance matrices of ranom effects. It is base on a general class of moels that inclues some of the existing moels as special cases an can be unerstoo as an EBLUP of the small area totals. Consequently, it oes not require the specification of a specific prior istribution for moel ranom effects. We also aim to evaluate this estimator in comparison with traitional synthetic an composite estimators that o not explicitly consier spatial variability. The paper is organize uner five sections. Section 1 introuces the context of the small area estimation an the goals of the paper. Section 2 reviews some traitional inirect estimators. Section 3 proposes an EBLUP estimator for small area totals base on spatial unit level ata. The estimator is assiste by a class of moels that fits into the general linear mixe theory. Section 4 escribes the esign of the Monte Carlo simulation stuy an presents empirical results. This stuy analyzes the performance of the propose estimator over the irect an inirect estimators using a real ata set from an agricultural survey conucte by the Portuguese Statistical Office. Discussion of the main finings of this stuy, along with some of its limitations an possible future evelopments are the subject of Section INDIRECT ESTIMATORS One possible approach for borrow information in the context of small area estimation base on implicit moels is to use irect moifie estimators. These estimators maintain certain esignbase properties such as approximately unbiase. This is the case of the regression estimator ([37]) (2.1) ˆτ,reg = ˆτ + (τ x ˆτ x ) ˆβ, = 1,..., D, where ˆτ is an estimator of the th omain total of the interest variable, usually the Horvitz Thompson or a poststratifie estimator an ˆτ x have the same
5 A Spatial Unit Level Moel for Small Area Estimation 159 meaning in relation to the vector of auxiliary variables x i = (x i1,..., x ip ), ˆβ = [ U i s υi 2 πi 1 x i x ] 1 i U i s υi 2 πi 1 x i y i are estimators of the regression coefficients obtaine using ata from the whole sample, υi 2 are regression weights an π i the inclusion probabilities resulting from the sampling esign. Estimator (2.1) is approximately unbiase, since E(ˆτ,reg ) = τ + τ x E(ˆβ) E(ˆτ xβ) τ (supposing E [ (ˆτ x τ x ) (ˆβ b) ] τ x b, where b = E(ˆβ) from the esignbase perspective 1 ). Although using information from outsie the omain for estimating the regression coefficients, usually these estimators still show low precision. An alternative is the synthetic estimation (whose properties epen on the assumptions of a postulate moel). From the esignbase point of view these estimators can be biase an inconsistent. A synthetic regression estimator can be presente as: (2.2) ˆτ,sreg = τ x ˆβ, = 1,..., D, where ˆβ is obtaine as before. A more extreme attitue uner a pure moelbase approach woul ignore the inclusion probabilities in estimating the regression parameters. The esignbase bias of estimator (2.2) is B(ˆτ,sreg ) τ x (b b ), assuming the regression weights are such that υi 2 p j=1 a j x ij, i U, where a j, j = 1,..., p, are arbitrary constants. This conition is always assure in the most typical situation of a nonweighte regression where the parameters υi 2 are assume constant. Further, typically estimator (2.2) has smaller variance than the irect moifie regression estimator but it is biase from the esignbase point of view. When ealing with small areas the reuction in variance associate with the synthetic estimator can be such that it will assure a mean square error (MSE) lower than the one obtaine through the use of the irect moifie estimator. There will always be a risk of a high bias an consequently the invaliity of any confience intervals obtaine uner repeate sampling. A significant avantage of synthetic estimation lies in the fact that it is always possible to obtain omain estimates, even in situations where the sample is very small or even zero. In orer to prevent the quality of the estimator being totally epenent on the postulate moel, some combine or composite estimators have been propose. A combine estimator typically presents the form of the weighte average of a esignbase estimator (approximately unbiase but with high variance) an a synthetic estimator (biase but with low variance): (2.3) ˆτ,com = λ ˆτ,es + (1 λ ) ˆτ,syn, = 1,..., D, with 0 λ 1. These estimators can be classifie in two main types (accoring to the way the weights λ are chosen): samplesize epenent weights an ata epenent weights. It is also possible to assume that the weights are chosen in 1 This conition supposes there is a sufficiently week correlation between ˆτ x an ˆβ what is usually easily achieve as ˆτ x an ˆβ are estimate at ifferent aggregation levels.
6 160 P.S. Coelho an L.N. Pereira a eterministic way using, for example, some previous knowlege or an informe guess. This woul result in what [39] call weights fixe in avance. A goo example of a combine regression estimator where the weights epen on sample size is the ampene regression estimator ([38]): (2.4) ˆτ,reg = λ ˆτ,reg + (1 λ ) ˆτ,sreg, = 1,..., D, where λ = 1 if ˆN N an λ = 0 otherwise, where N is the th omain population size an h is a positive constant. The authors suggeste to use h = 2. The basic iea for choosing h is to assure that the bias contribution from the synthetic component of the estimator is kept within acceptable limits. Another possible approach for borrow information in the context of small area estimation is to use a ataepenent combine estimator, through the moeling of the bias of the synthetic part of the estimator, thus proucing inirect estimates for the weights. Many of the moels that have been propose inclue ranom area effects an can be seen as particular cases of linear mixe moels. One of the best known moels applicable at unit level is the neste error regression moel ([8], [3]). All these approaches implicitly consier some kin of sectional correlation an the omain estimators are obtaine through EBLUP, empirical Bayes or hierarchical Bayes approaches. The wellknown neste error regression moel ([3]) has the form y i = x i β + u + ǫ i, = 1,..., D, i = 1,..., N, where u an ǫ are assume to be ii with zero means. It is also assume that u an ǫ i are mutually inepenent, V m (u ) = σ 2 u an V m (ǫ i ) = σ 2 υ 2 i where υ i are known constants. Here a common covariance between any two observations in the same small area is assume, Cov m (y i, y j ) = σ 2 u (i j). In this kin of moels it is assume that there is no sample selection bias, resulting that they are assume to hol both for the population an for the sample. This may be a very limiting assumption since small omain estimation is frequently neee in the context of informative sampling esigns. An alternative to the EBLUP in the context of informative sampling esigns is the pseuoeblup estimator ([29]). This estimator, base on the neste error regression moel, epens on survey weights an it is esignconsistent. 3. A COMBINED ESTIMATOR FOR SPATIAL DATA 3.1. A class of moels Let y i be the value of the interest variable for unit i (i = 1,..., n ) in small area ( = 1,..., D) an let x i = (x i1,..., x ip ) a vector of p unit level explana
7 A Spatial Unit Level Moel for Small Area Estimation 161 tory variables referring to the same unit. Consier the following class of moels: (3.1) y i = x i β+ H h=1 x (1)i q h,iu (1) h +x (2)i u(2) +ǫ i, =1,...,D, i =1,...,n, where β is a vector of p fixe effects; x (j)i is a vector of p j explanatory variables (typically a subvector of x i ) for the ith unit in small area ; q h,i are esign variables use to take into account the sampling esign an inicate that unit i belongs to a stratum or a sampling unit h (h = 1,..., H); u (1) h = col 1 j p 1 (u hj ) is a vector of p 1 ranom (or fixe) esign effects associate with stratum (or sampling unit) h; u (2) = col 1 j p2 (u j ) is a vector of p 2 ranom effects associate with omain ; l represents the geographical location associate to the centroi of omain ; f(l l e ) is a function of the vector l l e an ǫ i is the resiual term associate with unit i. We assume that E m u ( (1)) ( (2)) h = 0, Em u = 0, Em (ǫ i ) = 0, { ( (1) ) Σ E (1) {, h = g σ m u h u (1) g = 0, otherwise, E 2 m(ǫ i ǫ ej ) = i, = e, i = j 0, otherwise, E ( (2) ) m u u (2) e = { } Σ (2) f(l l e ), with Σ (1) = σ 2 (j, k =1,..., p U (1) 1 ), σ 2 = E(u jk U (1) hj u hk ), Σ (2) = { } jk σ 2 (j, k =1,..., p U (2) 2 ), an σ 2 = E(u jk U (2) j u k ). The moel is applie to ata jk from a sample of total size n = D =1 n, where n is the number of sampling units in area. It is also assume that ranom effects associate with ifferent aggregation levels are not correlate, E ( u (j) u (k)) = 0 for j k, an that the errors are noncorrelate with ranom effects, E ( u (j) ǫ ) = 0, j. In the propose moel, omain effects show a structure of spatial variability. The covariance between the ranom effects associate with omains an e epens on the vector efine by their geographical coorinates l l e. Some functions that can be applie to this context are presente in [28]. When the spatial covariance only epens on the istance l l e, then the function f( l l e ) is sai to be isotropic ([6]) an is typically such that lim l l e 0 f = 1. As the omains are not points in space, but areas, these coorinates are efine by their centrois. The assumption is that the lowest level of aggregation for which the georeferencing is available is the omain. Situations where the level of aggregation for which the referencing is available oes not coincie with the omains of stuy can generate special cases of moel (3.1). In particular, when georeferencing is possible at unit level, spatial variation can be moele through variancescovariances of the errors vector ǫ. Domain ranom effects represent the characteristics specific to the omain of stuy that affect the values of the interest variable an are not represente by the fixe effects at a higher level of aggregation. They can be thought of as moeling the bias of the synthetic part of the moel. Moreover, these omain ranom effects will now have the aitional role of bringing information from other omains, to explain the values of the interest variable in each omain.
8 162 P.S. Coelho an L.N. Pereira Design effects are use to take into account the sampling esign. The goal is to allow the moel to be applie to contexts with informative sampling esigns, overcoming the limitations ([27], [20]) of other ataepenent combine estimators that implicitly assume that the sampling esign is ignorable. The methoology propose can therefore be seen as moel assiste. The sample s is the result of a twostep proceure. First it is suppose that the finite population can be approximately escribe by a superpopulation moel. In the secon step it is assume that a sample is rawn from the finite population through a specific sampling esign. It is assume that the sample can be approximately escribe by moel (3.1), which has taken into account the existence of these two steps Estimation of moel parameters The moel 3.1 can be presente as a special case of the general linear mixe moel, grouping the unitspecific moels over the population: (3.2) y = Xβ + Zu + ǫ, where y is a vector of the target variable, X is a esign matrix of explanatory variables with rows given by x i, Z = [ ] Z (1) Z (2) is a esign matrix, u = ( col 1 j 2 u (j) ) is a vector of ranom effects an ǫ is a vector of errors. The covariance matrix of u is given by G = V m (u) = blockiag 1 j 2 G [ (j) ], where G (1) = { blockiag 1 h H Σ (1) } an G (2) =F Σ (2) with F = { f(l l e ) },,e = 1,..., D. Further, R = V m (ǫ) = iag 1 i n {σi 2 }, E m( u (1) ) ( =0, E m u (2) ) =0 an E m (ǫ) =0. 1 D Both covariance matrices G an R involve unknown variance components, represente by θ. Also the flexibility of the propose class of moels recommens proceeing in each application to the selection of a specific moel, i.e. to the choice of the explanatory variables an appropriate variancecovariance structures to u an ǫ. This step in moel selection an iagnosis is crucially important to obtaining a moel that can aequately escribe the behavior of the target population an can be performe as a systematic proceure like those propose by [7] or [43]. Once a specific moel has been selecte, the variance components θ nee to be estimate in orer to assess the variability of estimators or to preict the fixe an ranom effects. Several methos are available for estimating variance components, such as the analysis of variance (ANOVA) metho ([12]), the minimum norm quaratic unbiase estimation (MINQUE) metho ([33], [34], [35]) an the likelihoo methos. Some references about the maximum likelihoo estimation (MLE) metho ue to Fisher may be foun in [9], [1], [23], [21], [15] an [18]. On the other han, references about the resiual maximum likelihoo estimation
9 A Spatial Unit Level Moel for Small Area Estimation 163 (RMLE) metho propose by [41] an its extensions can be foun in [24], [10], [11], [2], [42], [14], [16], [17], among others. For etails about estimation general linear mixe moels see [19]. Now consier the ecomposition of all matrices into sample an nonsample components, where the subscript s is associate with the n sample units an r is associate with the (N n) nonsample units. The omission of the subscript inicates that the respective matrices allue to the whole population U s r. Assuming moel 3.2 hols an variance components are known, the best linear unbiase estimator of β an the best linear unbiase preictor of θ are given by (3.3) (3.4) β = ( X sv 1 s ũ = GZ sv 1 s ) 1 X s X s Vs 1 y s, ( ys X s β), where V s = E [ (y s X s β)(y s X s β) ] = Z s GZ s + R ss. The vectors u (j) may be preicte using ũ (j) = G (j) Z (j)s V 1 s (y s X s β), while the preictors of the errors ǫ, can be obtaine as ǫ = R s Vs 1 (y s X s β), where R s = [R ss R rs]. [ ] Rss R When the covariance matrix R = sr is blockiagonal, i.e. when R rs R rr there is no correlation between errors associate with the observations insie an outsie the sample, then R rs = 0 an ǫ r = 0. This is the case for moel (3.1). Nevertheless, it shoul be note that some situations can be evise, particularly when the spatial correlation can be establishe at unit level, where there is a correlation between moel errors that can be use in the preiction of ǫ r Estimation of omain totals The objective of the inference can be seen as to preict the total of an interest variable, τ, that uner the moel correspons to the summation of the realizations of the variable of interest over all the elements in the small area : (3.5) τ = i U y i = τ x, β + H h=1 τ x(1),h u(1) h + τ x(2), u(2) a + τ ǫ,, where τ ǫ, = i U ǫ i. It shoul be note that, from the moelbase point of view, (3.5) is a preictable function proucing inference in the narrow inference space [22]. An estimator for the small area total, τ, can be obtaine as (3.6) τ = 1 N ỹ = ỹ i = τ β x + v τsv 1 ( s ys X s β), i U where ỹ is the EBLUP of the vector y an v τs = τ z GZ s + 1 N R,s is the line vector of the moelbase covariances between the small area total τ an
10 164 P.S. Coelho an L.N. Pereira the observable vector y s, R,s = E(ǫ a ǫ s ), an 1 N is a unit vector of size N. It shoul be note that the estimator τ is the EBLUP of τ, given the observable ranom vector y s (cf. Appenix 1). When R,rs is a null matrix, then the EBLUP of the total is τ is given by a simplifie expression: (3.7) τ = Ẽ(τ,r u) = τ y,,s + τ x,,r β + τ z,,r GZ svs 1 ( ys X s β), where τ y,,s is the observe sample total in small area (cf. Appenix 2). It shoul also be note that many of the regression estimators that have been propose for small area estimation may be viewe as EBLUP of omain totals for particular cases of the class of moels (3.1). For instance, the form of the neste error regression moel an the ranom coefficient moel presente in Section 2 accor with class (3.1), with u (2) scalar, G (1) =0, G (2) = σu 2 I D an R = σ 2 I n. Also, the moel unerlying the synthetic regression estimator (2.2) is equivalent to consiering (3.1) with u (1) h scalar an taken as a fixe effect, G (2) =0 an R = σ 2 I n. Moreover, the irect moifie regression estimator (2.1), can be obtaine consiering u (1) h an u (2) scalars an taken as a fixe effects an R = σ 2 I n Domains not represente in the sample Situations may arise where some omains are not represente in the sample. If no sample falls into small area, then the respective ranom effects u (2) may still be preicte if there is covariance between u (2) an at least one of the small area ranom effects represente in the sample u (2) e (e = 1,..., D; e ). We have then (3.8) ũ (2) = G (2) (, Z (2)s V 1 s ys X s β), where G (2), = col ne 0 1 e D [ G (2),e ] [ = E u (2) u (2) ] (2) an G,e = Σ(2) f(l l e ). In an extreme situation where the small area effects u (2) are not correlate with any other small area effect for a omain represente in the sample, i.e. when G (2),e = 0, e : n e 0, then ũ (2) =0. The estimator τ is then reuce to a form similar to a following synthetic estimator: (3.9) τ = τ y,,s + τ x,,r β + H h=1 τ x(1),h,rũ(1) h. It may be note that this estimator may be written in the same generic form (3.10) ˆτ = τ x, ˆβ + f ( y s X s ˆβ),
11 A Spatial Unit Level Moel for Small Area Estimation 165 where f is use to weight the regression resiuals. This form puts in evience that estimator (3.9) can be seen as a combine estimator where the weights in f allow a correction of the synthetic part of the estimator τ x, β through the preiction errors in the omain that is the target of inference, but also in other omains spatially correlate. When no correlation between omains is specifie, the correction factor epens only on the preiction errors in the target small area an the estimator is reuce to a similar form to the ataepenent combine estimators presente in Section 2. These characteristics seem to be particularly interesting when estimating in small omains, where the available sample size is small, since it borrows information from outsie the omain of stuy in orer to assist the estimation. Moreover, taking avantage of the potential spatial correlation of ata it is possible to avoi the reuction of the propose estimators to pure synthetic estimators even when the sample size in the omain is null. 4. MONTE CARLO SIMULATION STUDY 4.1. Generation of the pseuopopulation For the simulation a pseuopopulation is use. This population is obtaine from a real ata set containing the responses to the 1993 wave of the Agricultural Structure Survey. It is an agricultural survey conucte by the Portuguese Statistical Office in the perio between agricultural censuses. The responses for the variable total prouction of cereals were extracte an circumscribe to the NUTSII of Alentejo. The total sample size in this region is 7,060 an the population size 47,049. The esign for the Agricultural Structure Survey is base on stratifie sampling. The sample is first stratifie using the região agrária as the level for geographic stratification. A região agrária is an aministrative ivision use for agricultural purposes. In each região agrária a new stratification is establishe base on Use Agricultural Surface (UAS) classes. In the same região agrária some other strata are efine base on the value of other variables consiere weakly correlate with UAS. In Alentejo there are 19 strata. For simulation purposes a pseuopopulation is generate by replicating the agricultural establishments in the sample proportionally to the inverse of their inclusion probabilities. The sampling frame resulting from this replication inclues the value of prouction of cereals for each establishment in 1993, an the same value reporte for 1989 (year of the agricultural census). The prouction in 1989 is use as an auxiliary variable in the moels use in the simulation. Also, geographical coorinates associate with the centrois of freguesias were recore.
12 166 P.S. Coelho an L.N. Pereira This was the lowest level of aggregation for which geographical referencing was available. This means that geographical ifferentiation between establishments inclue in the same freguesia is not available. A freguesia is an aministrative ivision that segments the Alentejo into 284 subregions Description of the simulations Using the sampling frame corresponing to the pseuopopulation of agricultural establishments we have run a Monte Carlo simulation. The goal is to evaluate the esignbase properties of a set of alternative estimators. Note that the approach followe in this paper is to evaluate the properties an relative merits of the propose estimators through simulation. In fact, ue to the complexity of these estimators their esignbase properties (e.g. bias, variance) are impossible to obtain through analytical methos. Also, their moelbase properties woul be of limite interest from the point of view of a benchmark with alternative irect estimators use in this simulation whose properties only make sense to evaluate from the esignbase perspective. The target parameter is the total of the variable prouction of cereals at freguesia level. The number of simulations performe is 560. In each simulation a sample is rawn from the pseuopopulation U, using a stratifie esign similar to the one use in the Agricultural Structures Survey. The only ifference in relation to that survey esign is that the sample size by stratum was reuce to 30% of the original size (2,118 establishments). The goal is to simulate a framework similar to that survey, but with a smaller sample size, enabling the evaluation of the estimators behavior in critical situations where the omain sample size is very small (sometimes only a few units or even none). This sampling esign leas to a relative precision of 7.5% (for a 95% confiencelevel) in the estimation of the total of the interest variable at the population level, using the Horvitz Thompson estimator. The expecte sample sizes for the 284 omains of interest vary from 0.3 to 45.8 units Estimators The estimators analyze in the simulation are presente in this section. They are mainly implementations of the irect, synthetic an combine regression estimators presente in Sections 2 an 3. It shoul be note that all the regression estimators inclue the same auxiliary variables (associate with the fixe effects), allowing a fair comparison of their relative merits. In what follows, the notation a is use to represent the small area of region a, where the regions correspon to the level of aggregation of NUTSIII an the small area of interest to freguesia. Table 1 summarizes the estimators use in the simulation.
13 A Spatial Unit Level Moel for Small Area Estimation 167 Table 1: Estimators use in the simulation stuy. Estimator Description ˆτ a1 Horvitz Thompson estimator ˆτ a2 Direct moifie regression estimator (2.1) ˆτ a3 Dampene regression estimator (2.4) ˆτ a4 ˆτ a5 τ a6 Pure synthetic regression estimator (2.2) with fixe effects estimate ignoring the sampling esign Synthetic regression estimator where the sampling esign is explicitly consiere through the inclusion of a vector of esign variables E hi inicating the belonging of each establishment i to the strata h =1,..., H Dataepenent combine regression estimator base on the neste error unit level regression moel τ a7 τ a8 Dataepenent combine regression estimator, similar to τ a6 but incluing fixe strata effects β h0, h =1,..., H Dataepenent combine regression estimator, base on a moel inclue in the propose class of moels (3.1), with ranom small area effects presenting a spatial covariance structure following an isotropic exponential moel. The isotropic exponential moel use to represent spatial variability in ˆτ a8 was suggeste in the moel iagnosis phase. We have teste several structures (exponential, spherical, linear, loglinear an Gaussian), through the evaluation the significance of covariance parameters (using Wal tests) an information criteria (such as AIC an BIC). Among the structures that showe statistical significance we retaine the one that minimize the several information criteria. Although we have chosen the exponential moel, some of the other structures resulte in very similar ajustments. Also note that for the ataepenent regression estimators the variance components are estimate through REML metho. The only exception regars estimator τ a8, where the parameter c e was estimate a priori through the ajustment of an exponential semivariogram to an empirical semivariogram. Note that the estimators inclue in the simulation vary in nature: ˆτ a1 an ˆτ a2 are esignbase estimators, ˆτ a4 an ˆτ a5 are synthetic estimators, while the others can be classifie as combine estimators as escribe in previous sections. The inclue estimators also ifferentiate in the way the sampling esign is (or not) taking into account: in ˆτ a1, ˆτ a2 an ˆτ a3 the sampling esign information is taking into account using sampling weights, ˆτ a5, ˆτ a7 an ˆτ a8 inclue fixe strata effects, while in ˆτ a4 an ˆτ a6 the sampling esign information is ignore.
14 168 P.S. Coelho an L.N. Pereira 4.4. Precision an bias measures The estimators uner consieration are evaluate using a set of precision an bias measures. In what follows K represents the number of simulations, an ˆτ k the th small area estimate of the total obtaine from the simulation k (k = 1,..., K) Unconitional analysis Taking into account the high number of small areas in the population (284) an in orer to facilitate the presentation of the simulation results, the small areas are ivie into six groups. Each group g contains D g small areas. Table2 presents the efinition of each group an the number of small areas involve. Table 2: Small area groups in the simulation stuy. Group Expecte sample size Number of small areas [0; 2] 20 2 [2; 3.5] 43 3 [3.5; 5] 49 4 [5; 10] 87 5 [10; + ] 65 Groups 1 to 5 were efine accoring to the expecte sample size of the small areas. Group 0 inclues small areas for which the total of the interest variable is zero (freguesias where there is no cereal prouction) regarless their size. The goal is to separate these small areas from the other groups to prevent them from changing the conclusions regaring the relative merits of the estimators. The Monte Carlo relative error for the estimators expecte value is on average 8.0% in group 1 an varies between 3.4% an 4.1% in groups 2 to 5. For the unconitional analysis the following measures were consiere for each group g: D g Average absolute bias: AAB g = Dg 1 AB, where AB = K 1 K ˆτ j τ ; Average MSE: AMSE g = D 1 g D g =1 j=1 MSE, where MSE = K 1 K (ˆτ j τ ) 2 ; =1 j=1
15 A Spatial Unit Level Moel for Small Area Estimation 169 Average variance: AV g = D 1 g D g V, where V = K 1 K (ˆτ j ˆτ ) 2 ; =1 Average absolutebiasratio: AABR g =D 1 g D g j=1 ABR, where ABR =AB / V ; Average coverage rate for a esignbase D g 100(1 α) confience interval: ACR g = Dg 1 TC, =1 where TC = 100 R /K an R represents the number of simulations for which the confience interval ˆτ j ± t α/2 V contains the true parameter τ. = Conitional analysis A conitional analysis was also conucte using a set of precision an bias measures for each small area. The superscript (n ) inicates that the respective measure is conitione to the realize sample size, n, in small area. They are: Conitional relative bias: CRB (n ) K n = Kn 1 (ˆτ j τ ) / τ ; j=1 Conitional relative stanar error: CRSE (n ) = Conitional variation coefficient: CVC (n ) = where V (n ) K n Kn 1 j=1 V (n ) / τ, (ˆτ j τ ) 2/ τ ; = K 1 (ˆτ j ˆτ ) 2 is the conitional variance ; n K n Conitional bias ratio: CBR (n ) = B (n / ) V n, where B (n K ) = Kn 1 n (ˆτ j τ ) is the conitional bias; Coverage rate of the conitional esignbase confience interval: CR (n ) = 100 R (n ) /K n, j=1 j=1 where R (n ) represents the number of simulations for which the confience interval ˆτ j ± t α/2 contains the true parameter τ. V (n )
16 170 P.S. Coelho an L.N. Pereira 4.5. Results Unconitional analysis Table 3 summarizes the unconitional results of the simulation stuy. The values for absolute bias, variance an MSE are presente relatively to the respective value associate with ˆτ a2. Table 3: Unconitional results. Group ˆτ a1 ˆτ a2 ˆτ a3 ˆτ a4 ˆτ a5 τ a6 τ a7 τ a8 Absolute bias Variance MSE Bias ratio Coverage Rate
17 A Spatial Unit Level Moel for Small Area Estimation 171 As expecte the two esignbase estimators ˆτ a1 an ˆτ a2 are approximately unbiase, even for very small areas. For these estimators the average coverage rate is very near the nominal confience level (95%). Nevertheless, especially in the smallest omains, they show variance an MSE substantially higher than those observe for the synthetic an combine estimators. It is also to be note that ˆτ a2 brings significant precision gains when compare with the Horvitz Thompson estimator. On the other han, the two synthetic estimators show very ifferent behavior. ˆτ a4, which can be viewe as a pure synthetic estimator shows isastrous behavior both in terms of bias an precision. These results are clear evience of the effects of ignoring an informative sampling esign. On the other han, the synthetic estimator with fixe strata effects, ˆτ a5, shows significant precision gains when compare to the irect regression estimator. These precision gains show a tenency to ecrease as the expecte sample size in the small areas increases. Its major rawback is relate to the high bias, which originates average bias ratios that are always above 2.6, compromising the construction of esignbase confience intervals. In fact the average coverage rate for this estimator is always below 0.46, an in many cases near The combine regression estimator with samplesize epenent weights, ˆτ a3, shows a systematic precision gain when compare to the irect regression estimator (with the exception of group 0, the ratio between the average MSE of the two estimators is between 0.79 an 0.87). Nevertheless, these gains are always moerate an substantially lower than the ones observe for the synthetic regression estimator. This estimator also shows a very goo behavior in what regars bias. In fact, although having an absolute bias higher than those observe for the irect estimators, the average bias ratio is always lower than 0.2, which originates an average coverage rate very near the nominal confience level. With regar to the combine estimators with ataepenent weights it can once again be observe that isregar of the sampling esign, as in estimator τ a6, prouces unesirable properties both in terms of bias an precision. In fact, τ a6 systematically shows higher MSE than the irect regression estimator (with the exception of group 1). It also exhibits ramatic biases (of the same magnitue as the synthetic estimator ˆτ a4 ) an bias ratios that on average are situate between 1.76 an The combine estimators τ a7 an τ a8, which explicitly consier strata effects, show very ifferent behavior. These estimators show average MSE that for the smallest small areas (groups 0 to 3) is near those observe for the best synthetic estimator, while allowing significant reuction in bias an bias ratio. In particular, τ a7 base on a neste error moel with fixe strata effects reveals important precision gains when compare to the irect regression estimator, ˆτ a2, or the samplesize epenent regression estimator, ˆτ a3, in groups 0 to 3.
18 172 P.S. Coelho an L.N. Pereira It is to be note that for groups 4 an 5 (corresponing to expecte sample sizes higher than 5 units) τ a7 shows some precision loss regaring the irect regression estimator, which is particularly important in group 5. It is also significant that, in the groups 0 to 3, τ a7 exhibits a MSE that is similar or even smaller than that associate with the synthetic estimator ˆτ a5. In these groups the increase in variance in τ a7 is more than compensate by the reuction in bias. In what regars bias measures, the estimator τ a7 shows a behavior that is situate between those recore for the irect an the synthetic estimators. The average absolute bias ratios vary from 0.40 to 2.58 (increasing with the reuction of the expecte sample size) an are strikingly lower than the ones associate with the synthetic estimators (varying between 15% an 50% of those obtaine for the best synthetic estimator). This results in average coverage rates for a esignbase confience interval that are substantially higher than those observe for the synthetic estimators. The estimator τ a8 consiers a spatial covariance base on an isotropic exponential structure at the small area level. For all small area groups (with the exception of group 0) it shows a small loss of precision when compare to τ a7, but still allows important precision gains regaring the irect estimators an the samplesize epenent regression estimator, ˆτ a3, in groups 0 to 3. It is worth noting that this ecline in precision is mainly inuce by an increase in variance, since τ a8 shows average absolute bias that is very near or even smaller than for τ a7 (mainly in the smaller areas). The bias ratios for τ a8 are, for groups 0 to 4, substantially smaller than those observe for τ a7, varying now from 0.42 to The reuction in the bias ratio tens to iminish with the increase in the expecte sample size, resulting that in group 5 the bias ratio of the two estimators is similar. In fact it is in smaller areas that τ a7 approximates more closely a synthetic estimator, allowing the aitional sample information use in τ a8 (from other spatially correlate small areas) to contribute to bias reuction. Between the combine estimators that allow precision gains in small areas groups 0 to 3, τ a8 is the one that shows the best behavior in terms of average bias ratio, which varies from 16% to 37% of those obtaine for the best synthetic estimator Conitional analysis Figure 1 an Table 4 summarize the simulation s conitional results for one small area in the stuy. Consiering the large number of small areas in the stuy (284), these ata are only intene to illustrate typical results associate with one of the smallest areas. The results refer to a small area with expecte sample size of 4.2 units, thus belonging to group 3.
19 A Spatial Unit Level Moel for Small Area Estimation 173 Figure 1: Conitional results. Table 4: Unconitional coverage rates. Sample size ˆτ a1 ˆτ a2 ˆτ a3 ˆτ a4 ˆτ a5 τ a6 τ a7 τ a The two esignbase estimators ˆτ a1 an ˆτ a2 show ba conitional properties, namely in what regars bias. In fact, both estimators show important conitional biases an bias ratios when the effective sample size eparts from the expecte sample size. This phenomenon is particularly notable in the Horvitz Thompson estimator. This bias tens to be negative for effective sample sizes that are smaller than expecte an positive for expecte sample sizes that are larger than expecte. Also, when effective sample size is smaller than the expecte sample size, the conitional variation coefficients ten to show a rising pattern with the increase in the effective sample size. The combine result of this bias an variance behavior is a conitional relative stanar error that for both estimators tens to increase with the effective sample size (for sample sizes above the expecte). When the effective sample size is significantly smaller or greater than the expecte sample size, these estimators (an mainly ˆτ a1 ) show a significant egraation in precision. These patterns are particularly notable in the small areas with a very small sample size.
20 174 P.S. Coelho an L.N. Pereira On the other han, although the synthetic estimators show very high bias an bias ratios (resulting in very low conitional coverage rates for a esignbase confience interval), they are seen to be approximately constant an therefore inepenent of the effective sample size for each small area. The conitional variation coefficient is clearly constant showing inepenence from the sample size in the small area. The combine result of these patterns is a relative stanar error which is also approximately invariant with the effective sample size. From this conitional point of view, the synthetic regression estimator can still be consiere one of the most precise estimators. For effective sample sizes that are smaller than the expecte sample size the combine regression estimator with samplesize epenent weights, τ a3, shows important conitional biases an bias ratios that for significant epartures are similar to those observe for the synthetic estimators. For sample sizes that are greater than expecte τ a3 tens to show a behavior similar to the irect regression estimator ˆτ a2. Therefore, the resulting conitional coverage rates for a esignbase confience interval also show a behavior similar to a synthetic estimator for sample sizes that are smaller than expecte an similar to the irect regression estimator when they are higher than expecte. The relative stanar error also tens to show the ba property observe for the irect estimator, characterize by an increase with the effective sample size, mainly for the smallest areas. The combine estimators τ a7 an τ a8 show interesting conitional properties as they show a mixe behavior between the irect regression estimator ˆτ a2 an the synthetic regression estimator ˆτ a5. This behavior is characterize by a significant resistance of precision an bias to epartures between the effective sample size an the expecte sample size. It can be observe that τ a7 shows a conitional bias an bias ratio that are approximately constant, although with a slight tenency to increase with the reuction of the effective sample size. As to bias, this estimator shows a clear avantage when compare to the synthetic estimators an even when compare to the irect estimators an the samplesize epenent combine estimator, particularly when the sample size eparts from the expecte sample size. This results in conitional coverage rates which in extreme situations are closer to the confience level than those associate with some esignbase estimators. The conitional variance only shows a very slight tenency to rise with an increase in the effective sample size, resulting in a conitional relative stanar error that is approximately constant. From the precision point of view it can be seen that this estimator is still competitive with the best synthetic estimator an maintains important precision gains when compare to the irect estimators an the samplesize epenent combine estimator (especially when the effective an expecte sample sizes are ifferent).
MSc. Econ: MATHEMATICAL STATISTICS, 1995 MAXIMUMLIKELIHOOD ESTIMATION
MAXIMUMLIKELIHOOD ESTIMATION The General Theory of ML Estimation In orer to erive an ML estimator, we are boun to make an assumption about the functional form of the istribution which generates the
More informationState of Louisiana Office of Information Technology. Change Management Plan
State of Louisiana Office of Information Technology Change Management Plan Table of Contents Change Management Overview Change Management Plan Key Consierations Organizational Transition Stages Change
More informationCrossOver Analysis Using TTests
Chapter 35 CrossOver Analysis Using ests Introuction his proceure analyzes ata from a twotreatment, twoperio (x) crossover esign. he response is assume to be a continuous ranom variable that follows
More informationData Center Power System Reliability Beyond the 9 s: A Practical Approach
Data Center Power System Reliability Beyon the 9 s: A Practical Approach Bill Brown, P.E., Square D Critical Power Competency Center. Abstract Reliability has always been the focus of missioncritical
More informationUCLA STAT 13 Introduction to Statistical Methods for the Life and Health Sciences. Chapter 9 Paired Data. Paired data. Paired data
UCLA STAT 3 Introuction to Statistical Methos for the Life an Health Sciences Instructor: Ivo Dinov, Asst. Prof. of Statistics an Neurology Chapter 9 Paire Data Teaching Assistants: Jacquelina Dacosta
More informationJON HOLTAN. if P&C Insurance Ltd., Oslo, Norway ABSTRACT
OPTIMAL INSURANCE COVERAGE UNDER BONUSMALUS CONTRACTS BY JON HOLTAN if P&C Insurance Lt., Oslo, Norway ABSTRACT The paper analyses the questions: Shoul or shoul not an iniviual buy insurance? An if so,
More informationDetecting Possibly Fraudulent or ErrorProne Survey Data Using Benford s Law
Detecting Possibly Frauulent or ErrorProne Survey Data Using Benfor s Law Davi Swanson, Moon Jung Cho, John Eltinge U.S. Bureau of Labor Statistics 2 Massachusetts Ave., NE, Room 3650, Washington, DC
More informationMASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 14 10/27/2008 MOMENT GENERATING FUNCTIONS
MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 14 10/27/2008 MOMENT GENERATING FUNCTIONS Contents 1. Moment generating functions 2. Sum of a ranom number of ranom variables 3. Transforms
More information10.2 Systems of Linear Equations: Matrices
SECTION 0.2 Systems of Linear Equations: Matrices 7 0.2 Systems of Linear Equations: Matrices OBJECTIVES Write the Augmente Matrix of a System of Linear Equations 2 Write the System from the Augmente Matrix
More informationENSURING POSITIVENESS OF THE SCALED DIFFERENCE CHISQUARE TEST STATISTIC ALBERT SATORRA UNIVERSITAT POMPEU FABRA UNIVERSITY OF CALIFORNIA
PSYCHOMETRIKA VOL. 75, NO. 2, 243 248 JUNE 200 DOI: 0.007/S336009935Y ENSURING POSITIVENESS OF THE SCALED DIFFERENCE CHISQUARE TEST STATISTIC ALBERT SATORRA UNIVERSITAT POMPEU FABRA PETER M. BENTLER
More informationStock Market Value Prediction Using Neural Networks
Stock Market Value Preiction Using Neural Networks Mahi Pakaman Naeini IT & Computer Engineering Department Islamic Aza University Paran Branch email: m.pakaman@ece.ut.ac.ir Hamireza Taremian Engineering
More informationINFLUENCE OF GPS TECHNOLOGY ON COST CONTROL AND MAINTENANCE OF VEHICLES
1 st Logistics International Conference Belgrae, Serbia 2830 November 2013 INFLUENCE OF GPS TECHNOLOGY ON COST CONTROL AND MAINTENANCE OF VEHICLES Goran N. Raoičić * University of Niš, Faculty of Mechanical
More informationMeasures of distance between samples: Euclidean
4 Chapter 4 Measures of istance between samples: Eucliean We will be talking a lot about istances in this book. The concept of istance between two samples or between two variables is funamental in multivariate
More informationCHAPTER 5 : CALCULUS
Dr Roger Ni (Queen Mary, University of Lonon)  5. CHAPTER 5 : CALCULUS Differentiation Introuction to Differentiation Calculus is a branch of mathematics which concerns itself with change. Irrespective
More informationA Case Study of Applying SOM in Market Segmentation of Automobile Insurance Customers
International Journal of Database Theory an Application, pp.2536 http://x.oi.org/10.14257/ijta.2014.7.1.03 A Case Stuy of Applying SOM in Market Segmentation of Automobile Insurance Customers Vahi Golmah
More informationDigital barrier option contract with exponential random time
IMA Journal of Applie Mathematics Avance Access publishe June 9, IMA Journal of Applie Mathematics ) Page of 9 oi:.93/imamat/hxs3 Digital barrier option contract with exponential ranom time Doobae Jun
More informationModelling and Resolving Software Dependencies
June 15, 2005 Abstract Many Linux istributions an other moern operating systems feature the explicit eclaration of (often complex) epenency relationships between the pieces of software
More informationSPECIFICS OF THE APPLICATIONS OF MULTIPLE REGRESSION MODEL IN THE ANALYSES OF THE EFFECTS OF GLOBAL FINANCIAL CRISES
Croatian Operational Research Review (CRORR), Vol., SPECIFICS OF THE APPLICATIONS OF MULTIPLE REGRESSION MODEL IN THE ANALYSES OF THE EFFECTS OF GLOBAL FINANCIAL CRISES Željko V. Račić Facult of Economics
More informationMath 230.01, Fall 2012: HW 1 Solutions
Math 3., Fall : HW Solutions Problem (p.9 #). Suppose a wor is picke at ranom from this sentence. Fin: a) the chance the wor has at least letters; SOLUTION: All wors are equally likely to be chosen. The
More informationOn Adaboost and Optimal Betting Strategies
On Aaboost an Optimal Betting Strategies Pasquale Malacaria 1 an Fabrizio Smerali 1 1 School of Electronic Engineering an Computer Science, Queen Mary University of Lonon, Lonon, UK Abstract We explore
More informationSampling design variance estimation of small area estimators in the Spanish Labour Force Survey
Statistics & Operations Research Transactions SORT 32 (2) JulyDecember 2008, 177198 ISSN: 16962281 www.iescat.net/sort Statistics & Operations Research c Institut Estaística e Transactions Catalunya
More informationLecture L253D Rigid Body Kinematics
J. Peraire, S. Winall 16.07 Dynamics Fall 2008 Version 2.0 Lecture L253D Rigi Boy Kinematics In this lecture, we consier the motion of a 3D rigi boy. We shall see that in the general threeimensional
More informationAchieving quality audio testing for mobile phones
Test & Measurement Achieving quality auio testing for mobile phones The auio capabilities of a cellular hanset provie the funamental interface between the user an the raio transceiver. Just as RF testing
More informationA New Evaluation Measure for Information Retrieval Systems
A New Evaluation Measure for Information Retrieval Systems Martin Mehlitz martin.mehlitz@ailabor.e Christian Bauckhage Deutsche Telekom Laboratories christian.bauckhage@telekom.e Jérôme Kunegis jerome.kunegis@ailabor.e
More informationCh 10. Arithmetic Average Options and Asian Opitons
Ch 10. Arithmetic Average Options an Asian Opitons I. Asian Option an the Analytic Pricing Formula II. Binomial Tree Moel to Price Average Options III. Combination of Arithmetic Average an Reset Options
More informationThe most common model to support workforce management of telephone call centers is
Designing a Call Center with Impatient Customers O. Garnett A. Manelbaum M. Reiman Davison Faculty of Inustrial Engineering an Management, Technion, Haifa 32000, Israel Davison Faculty of Inustrial Engineering
More informationPurpose of the Experiments. Principles and Error Analysis. ε 0 is the dielectric constant,ε 0. ε r. = 8.854 10 12 F/m is the permittivity of
Experiments with Parallel Plate Capacitors to Evaluate the Capacitance Calculation an Gauss Law in Electricity, an to Measure the Dielectric Constants of a Few Soli an Liqui Samples Table of Contents Purpose
More informationView Synthesis by Image Mapping and Interpolation
View Synthesis by Image Mapping an Interpolation Farris J. Halim Jesse S. Jin, School of Computer Science & Engineering, University of New South Wales Syney, NSW 05, Australia Basser epartment of Computer
More informationOptimal Control Policy of a Production and Inventory System for multiproduct in Segmented Market
RATIO MATHEMATICA 25 (2013), 29 46 ISSN:15927415 Optimal Control Policy of a Prouction an Inventory System for multiprouct in Segmente Market Kuleep Chauhary, Yogener Singh, P. C. Jha Department of Operational
More informationOpen World Face Recognition with Credibility and Confidence Measures
Open Worl Face Recognition with Creibility an Confience Measures Fayin Li an Harry Wechsler Department of Computer Science George Mason University Fairfax, VA 22030 {fli, wechsler}@cs.gmu.eu Abstract.
More informationForecasting and Staffing Call Centers with Multiple Interdependent Uncertain Arrival Streams
Forecasting an Staffing Call Centers with Multiple Interepenent Uncertain Arrival Streams Han Ye Department of Statistics an Operations Research, University of North Carolina, Chapel Hill, NC 27599, hanye@email.unc.eu
More information19.2. First Order Differential Equations. Introduction. Prerequisites. Learning Outcomes
First Orer Differential Equations 19.2 Introuction Separation of variables is a technique commonly use to solve first orer orinary ifferential equations. It is socalle because we rearrange the equation
More informationDEVELOPMENT OF A BRAKING MODEL FOR SPEED SUPERVISION SYSTEMS
DEVELOPMENT OF A BRAKING MODEL FOR SPEED SUPERVISION SYSTEMS Paolo Presciani*, Monica Malvezzi #, Giuseppe Luigi Bonacci +, Monica Balli + * FS Trenitalia Unità Tecnologie Materiale Rotabile Direzione
More informationThe oneyear nonlife insurance risk
The oneyear nonlife insurance risk Ohlsson, Esbjörn & Lauzeningks, Jan Abstract With few exceptions, the literature on nonlife insurance reserve risk has been evote to the ultimo risk, the risk in the
More informationMODELLING OF TWO STRATEGIES IN INVENTORY CONTROL SYSTEM WITH RANDOM LEAD TIME AND DEMAND
art I. robobabilystic Moels Computer Moelling an New echnologies 27 Vol. No. 23 ransport an elecommunication Institute omonosova iga V9 atvia MOEING OF WO AEGIE IN INVENOY CONO YEM WIH ANOM EA IME AN
More informationPerformance And Analysis Of Risk Assessment Methodologies In Information Security
International Journal of Computer Trens an Technology (IJCTT) volume 4 Issue 10 October 2013 Performance An Analysis Of Risk Assessment ologies In Information Security K.V.D.Kiran #1, Saikrishna Mukkamala
More informationThroughputScheduler: Learning to Schedule on Heterogeneous Hadoop Clusters
ThroughputScheuler: Learning to Scheule on Heterogeneous Haoop Clusters Shehar Gupta, Christian Fritz, Bob Price, Roger Hoover, an Johan e Kleer Palo Alto Research Center, Palo Alto, CA, USA {sgupta, cfritz,
More informationUnbalanced Power Flow Analysis in a Micro Grid
International Journal of Emerging Technology an Avance Engineering Unbalance Power Flow Analysis in a Micro Gri Thai Hau Vo 1, Mingyu Liao 2, Tianhui Liu 3, Anushree 4, Jayashri Ravishankar 5, Toan Phung
More informationTOMASZ KLIMANEK USING INDIRECT ESTIMATION WITH SPATIAL AUTOCORRELATION IN SOCIAL SURVEYS IN POLAND 1 1. BACKGROUND
PRZEGLĄD STATYSTYCZNY NUMER SPECJALNY 1 01 TOMASZ KLIMANEK USING INDIRECT ESTIMATION WITH SPATIAL AUTOCORRELATION IN SOCIAL SURVEYS IN POLAND 1 1. BACKGROUND First attempts at applying various approaches
More informationRisk Management for Derivatives
Risk Management or Derivatives he Greeks are coming the Greeks are coming! Managing risk is important to a large number o iniviuals an institutions he most unamental aspect o business is a process where
More informationRisk Adjustment for Poker Players
Risk Ajustment for Poker Players William Chin DePaul University, Chicago, Illinois Marc Ingenoso Conger Asset Management LLC, Chicago, Illinois September, 2006 Introuction In this article we consier risk
More informationISSN: 22773754 ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 3, Issue 12, June 2014
ISSN: 77754 ISO 900:008 Certifie International Journal of Engineering an Innovative echnology (IJEI) Volume, Issue, June 04 Manufacturing process with isruption uner Quaratic Deman for Deteriorating Inventory
More information9.3. Diffraction and Interference of Water Waves
Diffraction an Interference of Water Waves 9.3 Have you ever notice how people relaxing at the seashore spen so much of their time watching the ocean waves moving over the water, as they break repeately
More informationChapter 2 Review of Classical Action Principles
Chapter Review of Classical Action Principles This section grew out of lectures given by Schwinger at UCLA aroun 1974, which were substantially transforme into Chap. 8 of Classical Electroynamics (Schwinger
More informationRUNESTONE, an International Student Collaboration Project
RUNESTONE, an International Stuent Collaboration Project Mats Daniels 1, Marian Petre 2, Vicki Almstrum 3, Lars Asplun 1, Christina Björkman 1, Carl Erickson 4, Bruce Klein 4, an Mary Last 4 1 Department
More informationHull, Chapter 11 + Sections 17.1 and 17.2 Additional reference: John Cox and Mark Rubinstein, Options Markets, Chapter 5
Binomial Moel Hull, Chapter 11 + ections 17.1 an 17.2 Aitional reference: John Cox an Mark Rubinstein, Options Markets, Chapter 5 1. OnePerio Binomial Moel Creating synthetic options (replicating options)
More informationPatch Complexity, Finite Pixel Correlations and Optimal Denoising
Patch Complexity, Finite Pixel Correlations an Optimal Denoising Anat Levin Boaz Naler Freo Duran William T. Freeman Weizmann Institute MIT CSAIL Abstract. Image restoration tasks are illpose problems,
More informationFOURIER TRANSFORM TERENCE TAO
FOURIER TRANSFORM TERENCE TAO Very broaly speaking, the Fourier transform is a systematic way to ecompose generic functions into a superposition of symmetric functions. These symmetric functions are usually
More informationJitter effects on Analog to Digital and Digital to Analog Converters
Jitter effects on Analog to Digital an Digital to Analog Converters Jitter effects copyright 1999, 2000 Troisi Design Limite Jitter One of the significant problems in igital auio is clock jitter an its
More informationAnswers to the Practice Problems for Test 2
Answers to the Practice Problems for Test 2 Davi Murphy. Fin f (x) if it is known that x [f(2x)] = x2. By the chain rule, x [f(2x)] = f (2x) 2, so 2f (2x) = x 2. Hence f (2x) = x 2 /2, but the lefthan
More informationIntroduction to Integration Part 1: AntiDifferentiation
Mathematics Learning Centre Introuction to Integration Part : AntiDifferentiation Mary Barnes c 999 University of Syney Contents For Reference. Table of erivatives......2 New notation.... 2 Introuction
More informationA Theory of Exchange Rates and the Term Structure of Interest Rates
Review of Development Economics, 17(1), 74 87, 013 DOI:10.1111/roe.1016 A Theory of Exchange Rates an the Term Structure of Interest Rates HyoungSeok Lim an Masao Ogaki* Abstract This paper efines the
More informationFAST JOINING AND REPAIRING OF SANDWICH MATERIALS WITH DETACHABLE MECHANICAL CONNECTION TECHNOLOGY
FAST JOINING AND REPAIRING OF SANDWICH MATERIALS WITH DETACHABLE MECHANICAL CONNECTION TECHNOLOGY Jörg Felhusen an Sivakumara K. Krishnamoorthy RWTH Aachen University, Chair an Insitute for Engineering
More informationUnsteady Flow Visualization by Animating EvenlySpaced Streamlines
EUROGRAPHICS 2000 / M. Gross an F.R.A. Hopgoo Volume 19, (2000), Number 3 (Guest Eitors) Unsteay Flow Visualization by Animating EvenlySpace Bruno Jobar an Wilfri Lefer Université u Littoral Côte Opale,
More informationExploratory Optimal Latin Hypercube Designs for Computer Simulated Experiments
Thailan Statistician July 0; 9() : 793 http://statassoc.or.th Contribute paper Exploratory Optimal Latin Hypercube Designs for Computer Simulate Experiments Rachaaporn Timun [a,b] Anamai Nauom* [a,b]
More informationPerformance of 2D versus 3D Topographic Representations for Different Task Types
Performance of versus Topographic Representations for Different s Debra MacIvor Savage, Eric N. Wiebe an Hugh A. Devine North Carolina State University Raleigh, North Carolina In this stuy, a performance
More informationHeatAndMass Transfer Relationship to Determine Shear Stress in Tubular Membrane Systems Ratkovich, Nicolas Rios; Nopens, Ingmar
Aalborg Universitet HeatAnMass Transfer Relationship to Determine Shear Stress in Tubular Membrane Systems Ratkovich, Nicolas Rios; Nopens, Ingmar Publishe in: International Journal of Heat an Mass Transfer
More informationSensor Network Localization from Local Connectivity : Performance Analysis for the MDSMAP Algorithm
Sensor Network Localization from Local Connectivity : Performance Analysis for the MDSMAP Algorithm Sewoong Oh an Anrea Montanari Electrical Engineering an Statistics Department Stanfor University, Stanfor,
More informationFirewall Design: Consistency, Completeness, and Compactness
C IS COS YS TE MS Firewall Design: Consistency, Completeness, an Compactness Mohame G. Goua an XiangYang Alex Liu Department of Computer Sciences The University of Texas at Austin Austin, Texas 787121188,
More informationIf you have ever spoken with your grandparents about what their lives were like
CHAPTER 7 Economic Growth I: Capital Accumulation an Population Growth The question of growth is nothing new but a new isguise for an ageol issue, one which has always intrigue an preoccupie economics:
More informationOptimal Energy Commitments with Storage and Intermittent Supply
Submitte to Operations Research manuscript OPRE200909406 Optimal Energy Commitments with Storage an Intermittent Supply Jae Ho Kim Department of Electrical Engineering, Princeton University, Princeton,
More informationKater Pendulum. Introduction. It is wellknown result that the period T of a simple pendulum is given by. T = 2π
Kater Penulum ntrouction t is wellknown result that the perio of a simple penulum is given by π L g where L is the length. n principle, then, a penulum coul be use to measure g, the acceleration of gravity.
More informationLecture 17: Implicit differentiation
Lecture 7: Implicit ifferentiation Nathan Pflueger 8 October 203 Introuction Toay we iscuss a technique calle implicit ifferentiation, which provies a quicker an easier way to compute many erivatives we
More informationTO DETERMINE THE SHELF LIFE OF IBUPROFEN SOLUTION
TO DETERMINE THE SHELF LIFE OF IBUPROFEN SOLUTION AIM: To etermine the shelf life of solution using accelerate stability stuies. MATERIAL REQUIREMENT, ethanol, phenolphthalein, soium hyroxie (0.1N), glass
More informationSensitivity Analysis of Nonlinear Performance with Probability Distortion
Preprints of the 19th Worl Congress The International Feeration of Automatic Control Cape Town, South Africa. August 2429, 214 Sensitivity Analysis of Nonlinear Performance with Probability Distortion
More informationFROM THE EDITOR Challenges in Statistics Production for Domains and Small Areas (II) Other Articles
FROM THE EDITOR This is a secon issue evote to selecte papers presente at the International Conference (SAE005) on Challenges in Statistics Prouction for Domains an Small Areas. The Conference was hel
More information_Mankiw7e_CH07.qxp 3/2/09 9:40 PM Page 189 PART III. Growth Theory: The Economy in the Very Long Run
189220_Mankiw7e_CH07.qxp 3/2/09 9:40 PM Page 189 PART III Growth Theory: The Economy in the Very Long Run 189220_Mankiw7e_CH07.qxp 3/2/09 9:40 PM Page 190 189220_Mankiw7e_CH07.qxp 3/2/09 9:40 PM Page
More informationConsumer Referrals. Maria Arbatskaya and Hideo Konishi. October 28, 2014
Consumer Referrals Maria Arbatskaya an Hieo Konishi October 28, 2014 Abstract In many inustries, rms rewar their customers for making referrals. We analyze the optimal policy mix of price, avertising intensity,
More informationStress Concentration Factors of Various Adjacent Holes Configurations in a Spherical Pressure Vessel
5 th Australasian Congress on Applie Mechanics, ACAM 2007 1012 December 2007, Brisbane, Australia Stress Concentration Factors of Various Ajacent Holes Configurations in a Spherical Pressure Vessel Kh.
More informationAverageCase Analysis of a Nearest Neighbor Algorithm. Moett Field, CA USA. in accuracy. neighbor.
From Proceeings of the Thirteenth International Joint Conference on Articial Intelligence (1993). Chambery, France: Morgan Kaufmann AverageCase Analysis of a Nearest Neighbor Algorithm Pat Langley Learning
More informationMathematical Models of Therapeutical Actions Related to Tumour and Immune System Competition
Mathematical Moels of Therapeutical Actions Relate to Tumour an Immune System Competition Elena De Angelis (1 an PierreEmmanuel Jabin (2 (1 Dipartimento i Matematica, Politecnico i Torino Corso Duca egli
More informationThe Quick Calculus Tutorial
The Quick Calculus Tutorial This text is a quick introuction into Calculus ieas an techniques. It is esigne to help you if you take the Calculus base course Physics 211 at the same time with Calculus I,
More informationMathematics Review for Economists
Mathematics Review for Economists by John E. Floy University of Toronto May 9, 2013 This ocument presents a review of very basic mathematics for use by stuents who plan to stuy economics in grauate school
More information! # % & ( ) +,,),. / 0 1 2 % ( 345 6, & 7 8 4 8 & & &&3 6
! # % & ( ) +,,),. / 0 1 2 % ( 345 6, & 7 8 4 8 & & &&3 6 9 Quality signposting : the role of online information prescription in proviing patient information Liz Brewster & Barbara Sen Information School,
More informationModeling and Predicting Popularity Dynamics via Reinforced Poisson Processes
Proceeings of the TwentyEighth AAAI Conference on Artificial Intelligence Moeling an Preicting Popularity Dynamics via Reinforce Poisson Processes Huawei Shen 1, Dashun Wang 2, Chaoming Song 3, AlbertLászló
More informationStochastic Modeling of MEMS Inertial Sensors
BULGARIAN ACADEMY OF SCIENCES CYBERNETICS AND INFORMATION TECHNOLOGIES Volume 10, No Sofia 010 Stochastic Moeling of MEMS Inertial Sensors Petko Petkov, Tsonyo Slavov Department of Automatics, Technical
More information(We assume that x 2 IR n with n > m f g are twice continuously ierentiable functions with Lipschitz secon erivatives. The Lagrangian function `(x y) i
An Analysis of Newton's Metho for Equivalent Karush{Kuhn{Tucker Systems Lus N. Vicente January 25, 999 Abstract In this paper we analyze the application of Newton's metho to the solution of systems of
More informationCompact Form of Expressions for Inductance Calculation of Meander Inductors
SERBIAN JOURNAL OF ELECTRICAL ENGINEERING Vol. 1, No. 3, November 2004, 5768 Compact Form of Expressions for Inuctance Calculation of Meaner Inuctors Goran Stojanovic 1, Ljiljana Živanov 2, Mirjana Damjanovic
More informationA Generalization of Sauer s Lemma to Classes of LargeMargin Functions
A Generalization of Sauer s Lemma to Classes of LargeMargin Functions Joel Ratsaby University College Lonon Gower Street, Lonon WC1E 6BT, Unite Kingom J.Ratsaby@cs.ucl.ac.uk, WWW home page: http://www.cs.ucl.ac.uk/staff/j.ratsaby/
More informationProfessional Level Options Module, Paper P4(SGP)
Answers Professional Level Options Moule, Paper P4(SGP) Avance Financial Management (Singapore) December 2007 Answers Tutorial note: These moel answers are consierably longer an more etaile than woul be
More informationEmergence of heterogeneity in acute leukemias
Stiehl et al. Biology Direct (2016) 11:51 DOI 10.1186/s1306201601541 RESEARCH Open Access Emergence of heterogeneity in acute leukemias Thomas Stiehl 1,2,3*, Christoph Lutz 4 an Anna MarciniakCzochra
More informationRobust Reading of Ambiguous Writing
In submission; please o not istribute. Abstract A given entity, representing a person, a location or an organization, may be mentione in text in multiple, ambiguous ways. Unerstaning natural language requires
More informationAn intertemporal model of the real exchange rate, stock market, and international debt dynamics: policy simulations
This page may be remove to conceal the ientities of the authors An intertemporal moel of the real exchange rate, stock market, an international ebt ynamics: policy simulations Saziye Gazioglu an W. Davi
More informationImproving Direct Marketing Profitability with Neural Networks
Volume 9 o.5, September 011 Improving Direct Marketing Profitability with eural etworks Zaiyong Tang Salem State University Salem, MA 01970 ABSTRACT Data mining in irect marketing aims at ientifying the
More informationModelling football match results and the efficiency of fixedodds betting
Moelling football match results an the efficiency of fixeos betting John Goar * University of Wales Swansea Ioannis simakopoulos University of Wales Bangor * Corresponing author ress: Department of Economics
More informationTowards a Framework for Enterprise Architecture Frameworks Comparison and Selection
Towars a Framework for Enterprise Frameworks Comparison an Selection Saber Aballah Faculty of Computers an Information, Cairo University Saber_aballah@hotmail.com Abstract A number of Enterprise Frameworks
More informationAmerican Thoracic Society Documents
American Thoracic Society Documents An Official ATS Clinical Practice Guieline: Interpretation of Exhale Nitric Oxie Levels (FE NO ) for Clinical Applications Rae A. Dweik, Peter B. Boggs, Serpil C. Erzurum,
More informationUsing research evidence in mental health: userrating and focus group study of clinicians preferences for a new clinical questionanswering service
DOI: 10.1111/j.14711842.2008.00833.x Using research evience in mental health: userrating an focus group stuy of clinicians preferences for a new clinical questionanswering service Elizabeth A. Barley*,
More informationThe concept of onboard diagnostic system of working machine hydraulic system
Scientific Journals Maritime University of Szczecin Zeszyty Naukowe Akaemia Morska w Szczecinie 0, 3(04) z. pp. 8 90 0, 3(04) z. s. 8 90 The concept of onboar iagnostic of working machine hyraulic Leszek
More informationLagrangian and Hamiltonian Mechanics
Lagrangian an Hamiltonian Mechanics D.G. Simpson, Ph.D. Department of Physical Sciences an Engineering Prince George s Community College December 5, 007 Introuction In this course we have been stuying
More informationProduct Differentiation for SoftwareasaService Providers
University of Augsburg Prof. Dr. Hans Ulrich Buhl Research Center Finance & Information Management Department of Information Systems Engineering & Financial Management Discussion Paper WI99 Prouct Differentiation
More informationComparative study regarding the methods of interpolation
Recent Avances in Geoesy an Geomatics Engineering Comparative stuy regaring the methos of interpolation PAUL DANIEL DUMITRU, MARIN PLOPEANU, DRAGOS BADEA Department of Geoesy an Photogrammetry Faculty
More informationBond Calculator. Spreads (Gspread, Tspread) References and Contact details
Cbons.Ru Lt. irogovskaya nab., 21, St. etersburg hone: +7 (812) 3369721 http://www.cbonsgroup.com Bon Calculator Bon calculator is esigne to calculate analytical parameters use in assessment of bons.
More informationA Monte Carlo Simulation of Multivariate General
A Monte Carlo Simulation of Multivariate General Pareto Distribution an its Application 1 1 1 1 1 1 1 1 0 1 Luo Yao 1, Sui Danan, Wang Dongxiao,*, Zhou Zhenwei, He Weihong 1, Shi Hui 1 South China Sea
More information2 HYPERBOLIC FUNCTIONS
HYPERBOLIC FUNCTIONS Chapter Hyperbolic Functions Objectives After stuying this chapter you shoul unerstan what is meant by a hyperbolic function; be able to fin erivatives an integrals of hyperbolic functions;
More informationOptimizing Multiple Stock Trading Rules using Genetic Algorithms
Optimizing Multiple Stock Traing Rules using Genetic Algorithms Ariano Simões, Rui Neves, Nuno Horta Instituto as Telecomunicações, Instituto Superior Técnico Av. Rovisco Pais, 04000 Lisboa, Portugal.
More informationLearningBased Summarisation of XML Documents
LearningBase Summarisation of XML Documents Massih R. Amini Anastasios Tombros Nicolas Usunier Mounia Lalmas {name}@poleia.lip6.fr {first name}@cs.qmul.ac.uk University Pierre an Marie Curie Queen Mary,
More informationOption Pricing for Inventory Management and Control
Option Pricing for Inventory Management an Control Bryant Angelos, McKay Heasley, an Jeffrey Humpherys Abstract We explore the use of option contracts as a means of managing an controlling inventories
More informationHybrid Model Predictive Control Applied to ProductionInventory Systems
Preprint of paper to appear in the 18th IFAC Worl Congress, August 28  Sept. 2, 211, Milan, Italy Hybri Moel Preictive Control Applie to ProuctionInventory Systems Naresh N. Nanola Daniel E. Rivera Control
More informationFatigue design. environment neglect corrosion, wear, extreme events... approximate forces by a parametric desription or a synthetic spectrum
environment neglect corrosion, wear, extreme events... approximate forces by a parametric esription or a synthetic spectrum Uncertain atigue esign material ata CAtests external forces, time history CADrawing
More information