Methodology Atlas nº6. Atlas of variations in oncologic surgery hopitalisations
Methodology of the Atlas VPM 6METHODOLOGY OF THE ATLAS OF VARIATIONS INONCOLOGIC SURGERY HOSPITALISATIONThe basic architecture of variations measurementThe focus of analysis of geographic variations in practice is on the healthcareexposure experienced for a population living in a defined geographical area.Thus, the basic methodological approach to measuring these variations isecologic, and relies on a simple rate in which four elements should be defined:- the numerator (the clinical event of interest: hospital admission, surgical procedure, diagnosis test, etc.);- the denominator (population exposed -usually determined by the place of residence and the time period under scrutiny);- the “unit” of analysis (typically administrative health care demarcations, geographically defined around hospital services), and;- the timeframe (the period of time when the events in the numerator occur).These four elements allow the estimation of the average annual rate for theevent of study, along a period of time, in a certain area. Once the rate isestimated for each of the geographical areas under study, several analyses arecarried out:• Comparison of the rates with a view to detect significant variations. This can be done by estimating: a) sex and age-standardised rates which can then be compared across areas; b) statistics of variation based on the standardized rates (Extremal Quotient, Coefficient of Variation, etc.); c) statistics of variation based on expected cases (Standardized Utilization Ratios, Component of Systematic Variation, Empirical Bayes statistic, etc); and d) the relative variation between an event of interest and other one known as a low variation and non supply-sensitive procedure (e.g. hip-fracture repair).• A series of ecological secondary analyses aimed to assess eventual associations between the rates and certain demand variables (age, sex, deprivation, etc.), supply variables (beds, health care staff, tertiary care, teaching hospital, etc), and the effect of the Autonomous Community (AC) level, a proxy of regional healthcare policies in the Spanish National Health Service (SNS).The following paragraphs will review all those aspects describing the actualmethodology implemented in the production of the Atlas, specifically that for theAtlas of Variations in Oncologic Surgery across the National Health Service(SNS).11 Oliva G, Allepuz A, Kotzeva A, Tebé C, Bernal Delgado E, Peiró S, et al. Variaciones enhospitalizaciones por cirugía oncológica en el Sistema Nacional de Salud. Atlas Var Pract MedSist Nac Salud. 2009; 3(2):241-72.
The numerator: Oncologic surgery hospital admissionsA total of 9 oncologic conditions causing admission for surgical intervention inSNS hospitals were included in this Atlas: breast, bladder, colorectal, uterus,larynx, lung, stomach and oesophagus. The selection was based on theirimpact on morbidity and/or mortality in the Spanish population.The discharge dataset (hospital administrative database, CMBD) was thesource for case detection and inclusion, together with the case-day surgeryregistries (CMA) available in the 16 Autonomous Communities (AACC)participating in the project. The CMBD registers all hospitalisations in the SNShospital network though in the Basque country and Murcia it also includesprivate hospitals’ data. In turn, the Catalonian CMBD includes data from allhospitals integrated in the public utilisation network (Xarxa Hospitalariad’Utilització Pública – XHUP) regardless of public or private ownership. Allcases discharged between 2005 and 2007 were used in the analysis.CMBD and CMA provided the required clinical information (cause of admission,other diagnoses, surgical procedures) as well as administrative information(age, sex and address) for each of the cases. Diagnoses and procedures arecoded following the International Classification of Diseases 9th revision clinicallymodified (ICD9-CM). Case selection followed the criteria summarised in table 1.One more issue to be addressed in defining the numerator for the purpose ofthis Atlas, deals with the decision on using either patient or admissions incounting the total number of cases. In this case admissions had to be adoptedas basis for calculations; therefore variations in utilisation of surgical procedureseventually detected might be affected by existing variations in readmissionrates. As a result, the dominant pattern in the management of individualpatients should be included in the range of explanatory factors underlying thevariation found. An attempt to account for this effect was made by analysing theaverage admission rate for the same group of surgical procedures in eachhospital during the period of study.The denominator: exposed population2005 to 2007 census update (registered by the National Institute of Statistics,INE) was the source of information to determine the population under exposure.Municipalities’ population was disaggregated into 18 five-year age groups (from0 to 4 years old to 85 and over) and gender. The three census updates weremerged to build the person-time observation data. Some of the analyses werelimited to male, female or elderly population (aged 60 and over).The census basic information unit, municipality, creates some difficulties for theestimation of health area populations when it comes to cities which, due to theirsize and density, are divided into several healthcare demarcations. Thosehindrances were overcome by resorting to other sources of data such as localpopulation information systems or regional health identity card databases. Thisway, intra-municipality areas were set out when needed. 2
Methodology of the Atlas VPM 6 No exclusions have been applied for the denominator. It is worth noting that though the SNS’s wide coverage allows for equivalence between census and protected population, there is some degree of mismatching due to the people insured by public servants’ mutuality schemes (MUFACE, MUGEJU, ISFAS and others). This population is included in the denominator as part of the census figures, however, only the cases served in public hospitals would be included in the numerator. Since the majority of public servants opt for private provision, this factor might bias the rates due to the relatively different weight of civil servant populations across AACC and healthcare areas.Table 1. Inclusion and exclusion criteria applied in retrieving cases of oncologic surgery Main Diagnosis Procedures ObservationsBreast 174.*; 233.0 85.20; 85.21; 85.22; 85.23; 85.25; Only women cases of breast cancer. Excluded cases 85.33; 85.34; 85.35; 85.36; 85.41; referred to other centre 85.42; 85.43; 85.44; 85.45; 85.46; 85.47; 85.48;Colorectal 153.*; 154.0; 154.1; 45.72; 45.73; 45.74; 45.75; 45.76; Excluded cases referred to other centre 154.8; 230.3; 230.4; 45.79; 45.8; 46.10; 46.11; 46.13; Codes 46.1x y 46.2x only considered when coded 46.14; 46.20; 46.22; 46.23; 48.4*; together with the aforementioned codes (solo coding 48.5; 48.6*; Is referred to palliative interventions, excluded in this Atlas). 60.21; 60.29; 60.3; 60.4; 60.5;Prostate 185; 233.4; 60.61; Excluded cases referred to other centre 236.5 60.62; 60.69Uterus 179; 180.*; 182.*; 68.3; 68.4; 68.5*; 68.6; 68.7; 68.8; Excluded cases referred to other centre 233.1; 233.2; 68.9Stomach 151.*; 230.2 43.0; 43.4*; 43.5; 43.6; 43.7; 43.8*; Excluded cases referred to other centre 43.9; 43.1*; 44.3*Esophagus 150.* 230.1 42.4*; 42.32; 42.33; 42.39; 43.99; Excluded cases referred to other centre. 42.1*; 42.5 -42.69; 43.1* Codes 42.1*, 42.51-42.69 y 43.1* must go together with 42.4* (the opposite is referred to palliative intervention.)Larynx 161.*; 146.4; 148.2; 30.09; 30.1; 30.2*; 30.3; 30.4 Excluded cases referred to other centre. 230.0; 231.0 Code 31.1 (temporal tracheotomy) is not considered a palliative intervention; thus included Codes 30.3 and 30.4, permanent tracheotomy, is excluded, since is considered a palliative treatment.Lung 162.2 - 162.9; 32.28; 32.29; 32.3; 32.4; 32.5; 32.6; Excluded cases referred to other centre. 231.2 32.9; 34.01; 34.03; 34.04; 34.05; Trachea and pleura cancers are excluded from the 34.09 defintionBladder 188.*; 233.7 57.4*; 57.5*; 57.6; 57.7*; Excluded cases referred to other centre. 56.71 - 56.79. Codes 56.71 a 56.79 go together with radical cystectomy and are considered palliative care.[*] denotes the inclusion of all 4th and/or 5th digits following the root code.
The unit of analysis: healthcare areas180 geographical units make part of the analysis in this Atlas edition. Theyaccounted for an average annual population of 36,664,474 along the period ofstudy, according to the municipal census.The territorial division adopted in the Atlas follows the health circumscriptionscreated by each of the regional health administrations; these circumscriptionshold the added value of defining catchment areas for hospitals. Tables 2 and 3provide some of the relevant characteristics of these healthcare areas. Table 3also shows population size data for the age subgroup 60 and over; these dataare useful in describing the minimum size of geographical populations dealt within the analysis, an issue of interest in assessing the rates’ stability. Table 2. Some of the relevant characteristics of healthcare areas Innhabitants* CCAA Healthcare area Menor Mayor Andalucía 32 65.278 710.956 Aragón 8 55.560 377.592 Asturias 8 34.543 321.398 Baleares 3 80.912 747.728 Canarias Islas 7 10.078 796.933 Cantabria 3 86.238 299.729 Cataluña 37 3.713 1.562.825 Castilla la Mancha 8 79.473 398.443 Castilla León 11 91.364 354.788 Com. Aut. Vasca 7 203.690 419.983 Com. Valenciana 22 50.222 345.055 Extremadura 8 49.232 258.905 Galicia 16 33.000 487.281 La Rioja 1 286.981 286.981 Murcia 6 55.921 490.120 Navarra 3 61.900 427.503 (*) Average annual 2002-2004 4
Methodology of the Atlas VPM 6 Table 3. Some of the relevant characteristics of healthcare areas Women Men Aged over 60 Total Minimum 1.798 1.915 942 3.713 Percentile 5th 16.819 16.953 7.697 33.772 Percentile 25th 42.699 40.762 20.516 84.143 Percentile 50th 80.694 81.991 37.043 160.926 Percentile 75th 139.717 136.607 60.369 274.923 Percentile 95th 234.392 222.731 100.041 459.531 Maximum 825.542 737.283 413.953 1.562.825 Population 18.565.751 18.098.723 8.044.232 36.664.474 Average Population 2004-2006.Geographical assignment of casesAs explained above, the focus of the analysis is on the experiences of thepopulations residing in a well defined geographical unit during a certain periodof time. Thus, the analyses conducted compares the hospital admissionexperience of populations living in different territories, rather than hospitaladmission patterns, though they will obviously be closely interrelated. In thissense, cases assignment to the corresponding area becomes a central issue.Four AACC record address data following the INE’s municipalities coding(Catalonia, the Basque Country, Valencia and Galicia).The other 12 AACC usepostcodes to record residence data. Hence the vast majority of admissionscontained in the CMBD could be confidently assigned to a certain area.However, coding quality of this type of data (by any of the two procedures)varies widely with the percentage of records with incomplete address field -raising over 10% of the total in some CCAA. This circumstance has advised theimplementation of a set of assignment rules aimed to minimise the number ofrecords missed by the analysis:- Cases presenting complete address coding were assigned to the healthcare area that includes the corresponding municipality or postcode;- Cases with incomplete address coding but including at least the digits for province of the corresponding municipality or postcode system, were reassigned to the catchment area of the hospital where the patient was admitted, whenever the hospital was located in the province identified by the incomplete code;- Cases with incomplete address coding, where province digits did not match the hospital’s, were excluded from analysis.CMBD originated in certain CCAA deserve further consideration:
- Catalonia’s CMBD does not include the address code; instead, there is direct assignment of each discharge to a particular sector and healthcare area (GTS). Therefore, residents in other AACC, together with foreign residents and unknown addresses cannot be reassigned; - Navarra, Valencia and Murcia provided area assigned information (based on health identity cards) in addition to the address codes included in the CMBD database. Priority was given to the former, whenever inconsistencies between the two data sources were detected; - Murcia and Castilla-La Mancha contribute to the project AVPM-SNS with a restricted version of their CMBD which strictly includes the data required for the elaboration of each of the atlases, instead of the complete database submitted by the other CCAA. Though the assignment rules applied were identical, the results obtained from them might not be strictly comparable with those based on complete versions of the CMBD, as contributed by the rest of CCAA. Table 4 describes the results in assignment of cases to place of residence for the period 2002 to 2004 (the most recent where assignment was checked). The number of discharges with incomplete address code requiring reassignment ranged from 0.6% in Galicia to 48% in La Rioja (impeding disaggregation into its 2 health administrative circumscriptions). The ability to reassign patients – either directly or following the described set of rules- allowed reach very high values of assignment, between 93.2% in Cantabria and 99.3% in Galicia.Table 4. Assignment of cases to place of residence for the period 2002 to 2004CCAA Total Incomplete Total Unknown Residence Reasigned CMBD Residence Assigned n % n % n % %Andalucía 2.007.876 349.468 17,4 21.941 1,1 315.720 15,7 96,9Aragón 427.196 34.300 8 1.658 0,4 21.114 4,9 96,4Asturias 361.857 50.569 14 1.190 0,3 47.440 13,1 98,8Baleares 245.040 29.332 12 396 0,2 26.699 10,9 97,1Canarias 340.716 16.284 4,8 8.516 2,5 15.625 4,6 97,2Cantabria 155.567 35.733 23 968 0,6 29.032 18,7 93,2Cataluña 2.177.234 28.156 1,3 19.884 0,9 325 0 97,8Castilla la Mancha 109.728 4.972 4,5 201 0,2 2.533 2,3 97,6Castilla León 684.925 93.993 13,7 2.384 0,3 75.258 11 96,9Com. Aut. Vasca 711.768 25.424 3,6 4.354 0,6 9.892 1,4 97,1Com. Valenciana 1.248.945 32.924 2,6 14.250 1,1 9.666 0,8 96,6Extremadura 316.017 19.562 6,2 660 0,2 15.000 4,7 98,2Galicia 707.561 3.909 0,6 823 0,1 133 0 99,3La Rioja 80.389 38.893 48,4 271 0,3 35.572 44,2 95,5Murcia 62.981 1.121 1,8 1 0 23 0 98,3Navarra 191.201 2.867 1,5 3.552 1,9 195 0,1 96,7 6
Methodology of the Atlas VPM 6Rates and variation statisticsCrude, specific and standardised ratesCrude Rates are obtained by affiliating all discharges to the populations oforigin, and represent the number of discharges generated along the period2005-2007, over the annual average registered population multiplied by 3. As arule, all rates are calculated as per 10,000 inhabitants and year.2Since age and sex are potential determinants for morbidity, differences in ageand sex distribution across populations might explain differences in thecorresponding intervention rates. In order to control such effect, age and sex-standardized rates were calculated in each area by using the direct method,being 2001 Spanish population, the standard population of reference.Standardised rates would represent the rate that each of the areas would have -should all of them have the same age and sex structure of the Spanishpopulation. 95% confidence intervals were calculated for each of thestandardised rates.Expected cases and the standardised utilisation ratio (SUR)The statistical power of the calculations detailed above is subject to variationsdepending on the size of the populations and the number of interventions.Therefore, the error estimation linked to those rates can become significantwhen dealing with clinical procedures with low number of interventions and/orwith small populations of origin. In these situations a better option to be used isthe standardised utilisation ratio (SUR), a parameter quite similar to the wellknown standardised mortality ratio (SMR). Calculations require estimating thenumber of expected cases in each of the areas and contrasting (by using aratio) with the observed number of interventions.Expected cases are obtained by using the indirect method; thus applying theoverall standardised age and sex-specific rates to the corresponding sub-population in each area. Expected cases would represent the expectedutilisation if the different territories’ sub-populations homogenised theirhospitalisation levels according to overall rates. Unlike the direct method, SURdoes not allow for comparison across areas, because constant specific rates(those in the overall population) are applied onto the population pyramids ineach area, being needed to take into account differences in age and sexstructures across areas. However it allows for the comparison of each of theareas against a global pattern, in this case the overall population of the areasincluded in the study; therefore, it can be interpreted as a “relative risk”.2 It is important to keep in mind how discharges rather than persons are computed for thesecalculations; nevertheless, the equivalence discharge-person is expected to be high for themajority of the procedures under study –except bladder cancer surgery and to a lesser extentbreast and larynx cancer related surgeries. Re-interventions’ relevance (i.e. discharge-personequivalence) was assessed whenever it was from the same hospital fulfilling the criteria outlinedin table 1 (same tumour plus some of the corresponding surgical procedures) during the 3 yearsof the study period.
Statistics of variation3In addition to standardized rates and standardized ratios, Atlas VPM-SNSincludes the following statistics: 1. Ratio of Variation (Extremal Quotient, RV), as the ratio of the highest to the lowest observed rate values. This statistic is widely used and greatly valued due to its simplicity and intuitive interpretation (a RV of 2 denotes double utilisation). However, this indicator is quite limited due to its sensitivity to low rates, to differences in population size across areas, to readmissions and to extreme values; its statistical power is very low and should any of the areas involved present no interventions –which could often be the case in small area studies- the statistic will take infinity as a value. For all these reasons RV is usually substituted by the ratio of rates in areas in the 95th percentile of variation over those in the 5th percentile (RV95-5); this formulation reduces the impact of extreme values. It is often presented along with RV75-25 (ratio of percentile 75th to the 25th) which gives an idea of the variation within the 50% of the observations central in the distribution. 2. Coefficient of variation (CVu): standard deviation (Su) over the mean (Yu) CVu= Su/Yu Where: Su2 = Σ(Yi-Yu)2/(k-1); Su=√Σ(Yi-Yu)2/(k-1); and Yu= unweighted mean = ΣYi/k; where Yi = area i mean; k= number of areas The CVu expresses standard deviation (SD) in mean units; the advantage of it, compared to simple SD, is the independence from measure units. It can be interpreted as a relative variation (the higher the coefficient the more the variation). 3. Weighted Coefficient of Variation (CVw): standard deviation across areas (Sw) over mean across areas (Yw) , weighted by the size of each area. CVw=Sw/Yw where Sw2=Σ[ni(Yi-Yu)]/ (Σni-1); Sw=√[Σ[ni(Yi-Yu)]/ (Σni-1)] Yi=area i mean; Yu= population mean (=prevalence); Yw=Σni(Yi-Yu)/Σni= weighted mean; k=number of areas CVw is similar to the CVu, though it attaches more weight to the areas with larger population size and behaves better in the presence of substantial population size differences across areas.3 A detailed update on variation statistics can be accessed at: Ibañez B, Librero J, Bernal-Delgado E, Peiró S, González López-Valcárcel B, Martínez N, et al. Is there much variation invariation? Revisiting statistics of small area variation in health services research. BMC HealthServ Res. 2009;9:60. [Epub ahead of print]. (Acceded April 2011). Available at:http://www.biomedcentral.com/1472-6963/9/60 8
Methodology of the Atlas VPM 64. Systematic component of variation (SCV): It measures the variation in the deviation of the observed rate to the expected rate, expressed as a percentage of the latter. This measure derives from a model that acknowledges two sources of variation: systematic variation (among areas) and random variation (within each area). Expressed in mathematical terms [Σ((Oi-Ei)2/Ei2)-Σ(1/Ei)]/k where Oi= observed number of surgical procedures in area i Ei=expected number of surgical procedures in area i as a function of the age and sex population structure and the specific procedure rates by age and sex (indirect method adjustment) k= number of areas The higher the SCV value the larger the systematic variation (not attributable to randomness)5. Empirical Bayes Statistic (EB). As described some paragraphs above, the Standardised Utilization Ratio (SUR) is an estimator of each area’s “relative risk”; that is, the surgery utilisation risk compared to the group of reference (all areas) being highly dependent upon population sizes (its variance is inversely proportional to the expected number of cases). Extreme values of SUR are therefore the least precise estimations, since they correspond to the areas registering very low number of cases; however they may look dominant in the apparent geographical pattern. On the other hand the variability in the number of observed interventions is usually wider than the expected for a Poisson distribution (extra- variability). With a view to overcome all these issues derived from the direct use of SUR, several smoothing alternatives have been proposed in order to reduce the extra-variability. Based on the assumption of observed cases following a Poisson distribution, the empirical Bayesian method assumes SUR to be a random variable described by a log-normal distribution [log (SUR) ~N(m, s2)]. This model, known as mixed Poisson log-normal, is widely used in disease mapping literature, becoming the best fit (likelihood) estimation of the variance of the log-normal distribution for the SUR geographical pattern. EB is obtained by a quasi likelihood method. For empirical studies this statistic is extremely robust in detecting variability, showing stability across areas, particularly in low rates. Despite its scarcely intuitive interpretation (it should be interpreted in terms of relative variation) this is the most valuable statistic in appraising variability, and has been adopted as the variation reference pattern in the Atlas VPM- SNS publications.
6. Hip fracture repair´s SCV as a reference The hip fracture repair´s SCV is adopted as a criteria of need-based variability on the basis that variations in the utilisation of this procedure are unlikely to be affected by factors other than difference in incidence of femur fracture. A statistic can then be built as the ratio of a certain procedure SCV over femur fracture repair´s SCV (RSCVff). Obviously the value of RSCVff for femur fracture hospitalisation will be 1; an RSCVff of 2 for a procedure should be interpreted as the variability in the utilisation of the procedure doubling that registered for the femur fracture (i.e., based on need). 7. Autonomous Community effect Although healthcare area is considered the main unit of analysis, the specific way in what the SNS is organized, full devolution to the AACC with complete responsibility on regulation, funding and provision, requires analyze the relative effect of the AC level, in variation. Thus, Autonomous Community effect, considered as a proxy of regional policy effect, is studied by using Intraclass Correlation Coefficients or Rho statistics (as estimated by using multilevel analysis). Both of them draw out what portion of the variation is attributable to differences across regions, beyond the differences across areas. The higher the value, the more likely to signal AACC as a relevant underlying factor of variation (i.e. differences in healthcare policies might be considered as a associated factor)Exploring underlying factors in Atlas VPMAtlas VPM-SNS’ analyses are not only aimed to describe rates and theirvariation, but also to explore factors associated to the variability. In this sense,Atlas VPM-SNS bivariate and multivariate analyses are meant to seekassociation among supply or demand factors and the variation in rates.Supply factorsThe micro data from the National Statistic on Inpatient Regime Centres (EESRI)are the information source for variables regarding health services supply; dataare retrieved from the wave closest in time to the corresponding period ofanalysis. In the case of the Atlas on Oncological Surgery micro data from the2004 EESRI were pulled from the Ministry of Health, Social Policy and Equality(http://www.msc.es/estadEstudios/estadisticas/estadisticas/microdatos/frmBusquedaMicrodatos.jsp ).4,5 The availability of resources is then counted in thehealthcare area where each hospital is located. Typically, the levels ofresources are grouped into tertiles. The list of variables includes:4 Instituto de Información Sanitaria. Estadística de Establecimientos Sanitarios con Régimen deInternado. Handbook. (Accessed May 2009). Available at: http://www.msc.es/estadEstudios/estadisticas/estadisticas/microdatos/frmListadoMicrodatos.jsp5 Instituto de Información Sanitaria. Estadística de Establecimientos Sanitarios con Régimen deInternado. Questionnaire. (Accessed May 2009). Available at:http://www.msc.es/estadEstudios/estadisticas/estadisticas/microdatos/frmListadoMicrodatos.jsp 1
Methodology of the Atlas VPM 61. Hospital beds per 1,000 inhabitants. As functioning beds existing in the public network hospitals over the corresponding area population. Three categories matching the tertiles of the distribution are defined: 0.82-1.71, 1.72-2.42 and 2.43-6.15 beds per 1,000 inhabitants;2. Hospital size (beds). Using three categories: below 300 beds, between 300 and 599 beds and 600 beds or above;3. Internal Medical Residents (MIR). It allows characterising each hospital as postgraduate training and teaching centre. The variable splits hospitals into MIR training for any medical specialty or non-MIR training hospitals;4. Hospital doctors per 1,000 inhabitants. Specialists (any specialty) in the public network counted in full time equivalent (FTE); tertiles of the distribution yield the following three categories: from 0.06 to 1.05, from 1.06 to 1.30 and from 1.31 to 3.16 specialist doctors per 1,000 inhabitants;5. Surgeons per 10,000 inhabitants. It refers to surgeons in the public hospital network. The tertiles divided the distribution into below 0.15, from 0.15 to 0.40 and from 0.41 to 0.61 surgeons per 10,000 inhabitants;6. Surgical beds per 10,000 inhabitants. Functioning beds attached to general surgery and digestive tract services. The tertiles configure the following groups: below 0.38, from 0.39 to 0.56 and from 0.57 to 1.27 surgical beds per 10,000 habitants.Socio-economic factorsSocioeconomic variables are extracted from Spanish Economic Yearbookannually published by La Caixa6. This publication offers informationdisaggregated by municipality for those nodes over 1,000 inhabitants. In orderto build “average” values for each socioeconomic variable, and map them intothe healthcare areas, the figures for municipalities were aggregated andweighted by their population volume; this way, an average value was obtainedfor each healthcare area. As for the municipalities below 1,000 inhabitants, theywere assigned the average value of the health area where they were included(generally none of these municipalities represents more than 5-10% of anyhealthcare area population).On the other hand, some of the relevant economic variables may only beavailable with a number of years lag; for instance the 2006 yearbook offers asuite of 2006 data but when it comes to income it only shows 2003 data and2001 for educational level. All editions are searched to get data as closest tothe period of analysis as possible. The list of variables included: 1. Economic level 2003: available family income per inhabitant estimated by geographical area (municipal level) for year 2003. The Spanish Economic Yearbook defines 10 levels which for 2003 correspond to the following year income intervals: 1) <7,200 €; 2) 7,200-8,300; 3) 8,300- 9,300; 4) 9,300-10,200; 5) 10,200-11,300; 6) 11,300-12,100; 7) 12,100-6 Servicio de Estudios. Anuario Económico de España 2005. Barcelona: La Caixa;2005. (Acceso mayo 2009). Available at:http://www.lacaixa.comunicacions.com/se/index.php?idioma=esp
12,700; 8) 12,700-13,500; 9) 13,500-14-500 and 10) >14,500. Personal income available can be defined as the income on hand for household economies to expend and save, or rather the adding up of all effective revenue obtained by a household for a period of time. It could be considered as all revenue coming from salary, capital derived revenue, social benefits and cash transfers, deducting direct taxes and social security contributions by the family members discounted. For the Atlas, economic level was organised into tertiles: from 1.76 to 3.95, corresponding to areas of average available income below 10,000 € per inhabitant and year; from 3.98 to 6.46 corresponding to areas with an average available income between 10,000 and 11,700 € per inhab./year; and, from 6.47 to 9.40, areas with average available income between 11,700 and 14,500 or above per person and year2. Evolution of family income over the period 1999-2003. The Spanish Economic Yearbook estimates the evolution of available household income per municipality (period 1998-2003) in two steps. First the available household income for 1998 is estimated and then the coefficient of variation compared to 2003 estimations is calculated; once the variation in average household available income is obtained, intervals are defined corresponding to 10 different levels of household income variation: 1) below 10%; 2) 10 to 16%; 3) 16% - 21%; 4) 21% - 26%; 5) 26% - 34%; 6) 34% - 42%; 7) 42% - 50%; 8) 50% - 60%; 9) 60% - 72%; 10) More than 72%. Alternatively, the Atlas purposed tertiles are defined: from 2.75% to 5.00%; from 5.01% to 6.00%; and, from 6.01% to 10.3. Unemployment rate over total population in 2004. The Spanish Economic Yearbook defines it as the number of people registered in the employment bureau at each municipality by July the first 2004 with respect to the population of the municipality according to the census data by January first 2003 (registered unemployed*100/population). This unemployment rate displays by sex and age (16-24 years old; 25 to 49 years old and age 50 and over). The first choice for the denominator of this rate would be the active population in each municipality however the sample used for the national active population survey is not representative at municipal level rendering it useless for the level of disaggregation here required (it works fine for province and regional levels, though). Nevertheless, the unemployment rate obtained against the resident population is still a good indicator for comparison across municipalities. The distribution tertiles are: 0.90 to 2.90; 2.91 to 4.22; and 4.23 to 7.96.4. Unemployment rate over active population (24 to 49 years old). Similar to the previous variable but referred to the young adult population. Distribution tertiles group areas into the following intervals: 1.20 to 5.36; 5.42 to 8.00; and 8.02 to 14.255. Landline phones per 100 inhabitants by January 1st 2004. Distribution tertiles group areas into the following intervals: 23.9 to 32.8; 32.9 to 39.0; and, 39.1 to 61.1.6. Cars per 100 inhabitants. Automobile fleet (excluding vans and trucks) registered in each area by January 1st 2004. Distribution tertiles group 1
Methodology of the Atlas VPM 6 areas into the following intervals: 21.1 to 36.9; 37.0 to 42.1; and 42.2 to 62.4.7. Percentage of illiterates and without formal education in the year 2000. The variable takes as reference the household “head” level of education. Distribution tertiles group areas into the following intervals: 4.7% to 12.9%; 13.0% to 17.9%; and 18.0%-37.8%.8. Percentage of university level educated persons in the year 2000. Likewise the previous, the variable takes as reference the household “head” level of education. Distribution tertiles group areas into the following intervals: 4.7% to 9.0%; 9.1% to 12.4; and 12.4% to 25.7%.