Description

DEÜ MÜHENDİSLİK FAKÜLTESİ FEN VE MÜHENDİSLİK DERGİSİ Cilt: 9 Sayı: 3 sh Ekim 2007 APPLICATION OF A POPULATION BASED STUDY OF CORRESPONDENCE ANALYSIS IN CHOOSING A HEALTH INSTITUTION (SAĞLIK KURUMU

Information

Category:
## Government Documents

Publish on:

Views: 61 | Pages: 12

Extension: PDF | Download: 0

Share

Transcript

DEÜ MÜHENDİSLİK FAKÜLTESİ FEN VE MÜHENDİSLİK DERGİSİ Cilt: 9 Sayı: 3 sh Ekim 2007 APPLICATION OF A POPULATION BASED STUDY OF CORRESPONDENCE ANALYSIS IN CHOOSING A HEALTH INSTITUTION (SAĞLIK KURUMU SEÇİMİNDE UYGUNLUK ANALİZİNİN TOPLUM TABANLI BİR ÇALIŞMAYA UYGULANMASI) Aslı SÜNER*, Can Cengiz ÇELİKOĞLU* ABSTRACT/ÖZET Correspondence analysis is a method that makes it easy to interpret the categorical variables given in contingency tables, showing the similarities between row and column variables and associations as well as divergences among these variables via graphics on a lower dimensional space. It factors the data matrix into the rows and columns, shows each of them with a separate graphics; it releases important information about the structure of the data. Having lots of categorical data in applied science, such as in medicine, biology and economy, makes correspondence analysis a very popular method. In correspondence analysis, the aim is to explain the associations among the variables on a lower dimensional space. A multidimensional graphic consisting the points regarding the levels of categorical variables is drawn using mathematical methods to determine the structure of the contingency table. There are similarities between this analysis and other multivariate methods; mainly Principal Component Analysis, Log-Linear Models and Multi-dimensional scaling. In this study, correspondence analysis will be applied to the epidemiological research data obtained from the DEU School of Medicine Department of Public Health. Uygunluk Analizi, kategorik değişkenlerin yorumlanmasını kolaylaştıran, çapraz tablolarda satır ve sütun değişkenleri arasındaki benzerlikleri ve bu değişkenlerin birlikte değişimlerini, daha az boyutlu bir uzayda grafiksel olarak gösteren bir yöntemdir. Bu teknik, veri matrisinin satır ve sütun bölgelerine ayrıştırılması üzerine yoğunlaşıp, elde edilen bileşenleri ayrı ayrı grafiklerle göstererek, veri setinin yapısına ilişkin önemli bilgiler vermektedir. Özellikle tıp, sağlık bilimleri, biyometri, ekonomi, pazarlama ve sosyal bilimler gibi kategorik verilerin analizine ihtiyaç duyulan alanlarda oldukça popüler bir yöntemdir. Uygunluk analizinde, değişkenler arasındaki ilişkilerin indirgenmiş boyutlu bir uzayda sunulması amaçlanmaktadır. Bu analizde, çapraz tabloların yapılarını belirlemek amacıyla matematiksel teknikleri kullanarak çok boyutlu uzayda değişkenlerin kategorilerini temsil eden noktaları içeren bir grafik oluşturulur. Uygunluk analizinin, çok değişkenli analiz tekniklerinden Temel Bileşenler Analizi, Log-Doğrusal ve Çok Boyutlu Ölçekleme yöntemleriyle de benzerliği bulunmaktadır. Bu çalışmada, DEÜ Tıp Fakültesi Halk Sağlığı Anabilim Dalında yürütülen araştırma verileri kullanılarak, uygunluk analizi ile ilgili bir uygulama yapılmıştır. KEYWORDS/ANAHTAR KELİMELER Correspondence analysis, Multivariate analysis methods, Analysis of categorical data Uygunluk analizi, Uyum analizi, Çok değişkenli analiz yöntemleri, Kategorik veri analizi * Dokuz Eylul University, Faculty of Art and Science, Department of Statistics, İZMİR Sayfa No: 2 A. SÜNER, C. C. ÇELİKOĞLU 1. INTRODUCTION In scientific studies depending on real life applications; the features of the objects under observation are related to each other since the events that are being discussed are usually influenced by many factors. It is an obligation to evaluate the objects under observation, from all possible aspects, to get valid and reliable results from the studies (Tatlıdil, 1996). So as to study on multivariate data and analysis, it is referred to the multivariate statistical methods. Reduction of difficulties confronted in the commendation and summarization of the results related to the multiple variables is aimed with the help of multivariate statistical methods used in the analysis of large data matrix (Burcu and Çetin, 2004). Methods with one variable have a restricting assumption; they require experimental control of many factors in the event (and necessity) of observing the effects of each factor at one time (Çetin, 2003). On the other hand, in multivariate statistical methods, get the event analysis integrated and to explain the collinearity of the variables supplying this integrity is aimed. The multivariate statistical methods are divided into two according to their interrelation as; methods used in the analysis of dependency structures and methods used in the analysis of interdependency structures. In the methods used in the analysis of dependency structures, one or more variable must be dependent on the other variables and their values must be predictable and explanatory (Bayram, 2000). In the methods used in the analysis of interdependency structures, there is the condition that the variables would not be explained with the help of other variables, the values would not be predicted or the variables would not be defined as dependent or independent, and the whole relation between the variables are considered (Bayram, 1996). The general tendency of where and when the multivariate statistical methods are used can be classified as in the Table 1 (Çetin, 2003). Basic multivariate statistical methods are explained and their classification is defined as is shown in Figure 1 (Hair et al., 1992). Apart from many other multivariate statistical methods; in correspondence analysis, not only the inner relations among the levels of the rows and the columns in data matrix are given, but also their similarities and the differences are shown. Moreover, the data structure is interpreted by observing the levels of the variables, which state the features expressed in rows and columns (Behdioğlu, 2000). Correspondence analysis is similar to principal component analysis. In both methods the dimension of the data matrix is aimed to be reduced and presented in a simple way in multidimensional space; the data is analyzed in reduced dimensions with the help of singular value decomposition (Etikan et al., 2000). Similar to the principal component analysis decomposing of general change into units, data can be divided according to inertia in correspondence analysis. There is a difference between these two techniques on type of the data matrix. In principal component analysis, the data consist of continuous or discrete measured variables, which supply a multivariate normal distribution. In the other hand in correspondence analysis the data are categorical and there is no need for them to be distributed, according to a distribution assumption (Etikan et al., 2000). The non-linear method is neither wanted nor needed in correspondence analysis (Seyfullahoğulları, 2002). Correspondence analysis has some similarities to the Log-linear analysis and multidimensional scaling. But correspondence analysis differs from log-linear models since it does not need an assumption related to the distribution and differs from multidimensional scaling in the base of showing the relations between the categories and variables in the same space (Tuna and Kıroğlu, 1996). Fen ve Mühendislik Dergisi Cilt : 9 Sayı : 3 Sayfa No: 3 Multivariate Method Factor Analysis Cluster Analysis Discriminant Analysis Multivariate Regression Analysis Multidimensional Scaling Method Correspondence Analysis Principal Component Analysis Logistic Regression Analysis Canonical Correlation Analysis Table 1. The general tendency of multivariate statistical methods When It Is Used? It is used to find and discover a less number of meaningful variables (factors) by gathering a large number of variables. It is used to classify the grouped data and to get proper, useful, summarized data. It is used to determine the discriminant factors effecting the discriminant of groups and to assign an object, whose source is not known, to a group. It is used to observe and analyze the relationship between a dependent variable and one or more independent variables. It is used in the analysis of behavioural data such as personal choices, manners, tendencies, beliefs and preferences. It is used in the analysis of the categorical data. The graphical outcome of the analysis has rich data that can be used for decision making. It is used to reduce the data, to re-form the data set in order to be analyzed by some methods, to calculate the principal component scores and to order the units according to these scores. It is a method used when the dependent variable is a discrete variable composed of binary (0,1) or multi levels It is an expansion and extension of the multiple regression analysis. Function It explains the inner relations between the multivariate variables and finds the common dimensions (factors) standing in the centre of the variables. It classifies the entities (individuals or objects) to smaller sub-groups by their similarities. It predicts the group differences and an entity s (an individual or an object) belonging to a group or a cluster defined by metric independent variables. It predicts the changes of dependent variables as a response to independent variables. It gives the structure of the objects as similar as possible to the original vision with as few dimensions as possible. It visually presents the closeness and similarities between the objects and gives the relations, which were not given in the tables. It gives a chance to find out the relations, which were not reached before, and finding unusual results. It is a method to group the observations. It relates various independent and dependent metric variables. Correspondence analysis can be applied in two ways according to the number of variables and dimensions in the cross tables (Cangür et al., 2005). In its very simple form, it is called Simple Correspondence Analysis and is used in examination of two sided cross tables. When the number of variables is not limited, the variables are coded as a matrix and applied to the multi-way cross table, the case is called as a Multiple Correspondence Analysis (Greenacre and Hastie, 1987). Multiple correspondence analysis is also known as Homogeneity Analysis, HOMALS (Homogeneity analysis by Alternating Least Squares) and it is named in America as optimal scaling, optimal scoring, reciprocal averaging and appropriate scoring ; in Japan as quantification methods ; in Holland as homogeneity analysis ; in Canada as dual scaling and in Israel as scalogram (Gifi, 1990; Etikan et al., 2000; Bayram, 2000). In Turkish literature, these are used as Uyum Analizi, Karşılık Getirme Analizi and Görsel İlişki Analizi (Akıncı and Atılgan, 2005). Sayfa No: 4 A. SÜNER, C. C. ÇELİKOĞLU Figure 1. Selecting a multivariate technique Fen ve Mühendislik Dergisi Cilt : 9 Sayı : 3 Sayfa No: 5 The relationships among categorical variables in large tables can be summarily described with correspondence analysis. The associations between rows and columns in a frequency table are mapped graphically, as points in a space of few dimensions. Conceptually, the analysis steps are simple. Category profiles and marginal proportions are computed, the distances between these points are calculated, and the best-fitting space of n dimensions is located. Correspondence analysis is preferred since it is a data reduction and residual analysis and it gives information about two or more dimensional graphs and categorical variables, since it has the ability to explain the lack of homogeneity in row profiles or the (dependency or interaction) between the rows and columns of the cross tables in a fewer dimensions (Cangür et al.., 2005). Since the studies on categorical data analysis high in number, the use of correspondence analysis is popular. It is a widespread method in the fields where the categorical data analysis is needed, such as medicine, medical sciences, biometry, economy, marketing and social sciences. Because it is useful in the analysis of cross tables and it presents easy, understandable and easily interpretable graphical visualizations, this method has gained a place in marketing research lessons course plans at universities. Especially, related to the increasing rate of computer use, it has become an optimal scaling method which can be found in statistical packages such as SPSS, MINITAB and SAS (Clausen, 1998). 2. APPLICATION In this research, data concerning adults who reside in Narlıdere, İzmir were examined in detail. The utilized data belongs to 348 people, who have illnesses, and was gathered for adults mental health research at Narlıdere education and research area by the DEU School of Medicine Department of Public Health. Application of a population based study of correspondence analysis in choosing a health institution with these data was aimed. Correspondence analysis was carried out so as to analyze the medical establishment, consulted by people who reside in Narlıdere according to their disease group, their age classification and the health insurance they have. This data set contains information on the characteristics of ages, health insurance, health institute and disease. The following tables show the variables, along with their variable labels, and the value labels assigned to the categories of each variable in the data set. Distribution of categories of age is given in Table 2. Looking at this table, 39.7% of patients (138 people) are in age group, 27.0% (94 people) are in age group, 21.6% (75 people) are at 65 or above and 11.8% (41 people) are in age group. Distribution of the categories of age group variable can be seen in Figure 2. Table 2. The frequency distribution of age variable Age Frequency Percent Cumulative Percent ,7 11, ,0 38, ,7 78, ,6 100,0 TOTAL ,0 Sayfa No: 6 A. SÜNER, C. C. ÇELİKOĞLU AGE ,6% ,7% ,7% ,0% Figure 2. Pie chart of age groups When looking at Table 3, distribution of categories of patients health insurance variable can be seen. In this table, 38.8% of patients (135 people) have health insurance from social security organization, 36.5% (127 people) have health insurance from retirement fund, 10.1% (35 people) have health insurance from Bağkur, 3.7% (13 people) have green card coverage and 2.3% have commercial health insurance while 8.6% (30 people) do not have any health insurance. Distribution of the categories of patients health insurance variable is illustrated in Figure 3. Table 3. The frequency distribution of health insurance variable Health Insurance Frequency Percent Cumulative Percent None 30 8,6 8,6 Social security organization ,8 47,4 Retirement fund ,5 83,9 Bağkur 35 10,1 94,0 Commercial 8 2,3 96,3 Green card coverage 13 3,7 100,0 Total ,0 Frequency None Soc. Sec. Ret. Fund Bagkur Com. Green card coverage Health Insurance Figure 3. Bar chart of health insurances Fen ve Mühendislik Dergisi Cilt : 9 Sayı : 3 Sayfa No: 7 Distribution of categories of patients referred health institute variable is given in Table 4. In this table, 25.3% of patients (88 people) enter university hospitals, 20.1% (70 people) enter social insurance institute hospitals, 9.2% (32 people) enter public hospitals, 4.6% (16 people) enter health unit and 4.0% (14 people) enter private hospitals. It is unknown which medical establishment 36.8% of patients (128 people) enter. Frequency distribution of health institute variable is examined in detail in Figure 4. Table 4. The frequency distribution of health institute variable Health Institute Frequency Percent Cumulative Percent Health unit 16 4,6 4,6 Social insurance inst ,1 24,7 Public 32 9,2 33,9 University 88 25,3 59,2 Private 14 4,0 63,2 Unknown ,8 100,0 TOTAL , Frequency H. Unit Soc. Inst. Public University Private Unknown Health Institute Figure 4. Bar chart of health institutes When looking at Table 5, distribution of categories of disease variable is seen. In this table, 27.0% of patients (94 people) have cardiovascular diseases (hypertension, heart attack, heart failure and apoplexy), 15.8% (55 people) have both cardiovascular and other diseases except diabetes, 14.1% (49 people) have rheumatism, 12.6% (44 people) have other diseases (tuberculosis, epilepsy, having obstacles, cancer, etc.), 8.0% (28 people) have diabetes, 7.8% (27 people) have gut-rot, 7.5% o (26 people) cardiovascular disease, diabetes and any other diseases, 4.0% (14 people) have asthma, 3.2% (11 people) have depression. Distribution of the categories of the disease variable can be evidently seen in Figure 5. Sayfa No: 8 A. SÜNER, C. C. ÇELİKOĞLU Table 5. The frequency distribution of disease variable Disease Frequency Percent Cumulative Percent Cardiovascular 94 27,0 27,0 Diabetes 28 8,0 35,1 Asthma 14 4,0 39,1 Gut-rot 27 7,8 46,8 Rheumatism 49 14,1 60,9 Depression 11 3,2 64,1 Other 44 12,6 76,7 Cardiovascular, Diabetes and Any Other 26 7,5 84,2 Cardiovascular and Any Other (except diabetes) 55 15,8 100,0 Total ,0 Frequency CV Diabts Asthma Gut-rot Rheu. Dep. Other CV+D+O CV+O (- D) Health Institute Figure 5. Bar chart of diseases 3. RESULTS OF ANALYSIS In the application, medical establishment consulted according to disease type was examined thoroughly by simple correspondence analysis as a sample in the first place. Distribution of age group, health insurance and disease type according to medical establishment were studied by multiple correspondence analysis. The symmetrical normalization makes it easy to examine the relationship between referred health institute of patient and diseases of patients categories. For example, diabetes are near the social insurance institute hospital category, while other diseases are closest to health unit. Cardiovascular diseases and other diseases seem to be associated with public hospitals; gutrot and depression are not strongly associated with any particular referred health institute of patient category. This relationship is shown in Figure 6. Fen ve Mühendislik Dergisi Cilt : 9 Sayı : 3 Sayfa No: 9 At the graphic obtained as a result of multiple correspondence analysis in Figure 7, it was seen that medical establishments consulted by adults, who are at 34 and under, having gut-rot disease and not having any health insurance was unknown. It is seen that individuals, who are at age group and having rheumatism and asthma, prefer going private hospitals and individuals, who have diabetes and social insurance institute health insurance, prefer going to social insurance institute hospitals. It was seen that adults, who are at least 50 and above and have cardiovascular disease (hypertension, heart attack, heart failure and apoplexy) and other diseases and have retirement fund health insurance or private health insurance, prefer going to university hospitals. Besides, individuals who have other diseases (tuberculosis, epilepsy, having obstacles, cancer, etc.) and have Bağkur health insurance prefer going to public hospitals and health unit. When analyzing the graphic above, it is seen that categories of individuals, who have green card coverage, depression and cardiovascular, diabetes and other diseases, is distant from origin. This denotes that marginal frequencies of these categories are less then the others. Figure 6. The symmetrical normalization graph of disease and health institute Sayfa No: 10 A. SÜNER, C. C. ÇELİKOĞLU Figure 7. Discriminant measures graph of variables 4. CONCLUSIONS Correspondence analysis is one of the multivariate statistical methods, in which the corresponding relationships between categorical variables. This analysis, which started to be used after 1930's, differs from other types of multivariate analyses, such that, it does not only help investigating the corresponding relationships defined in the columns or rows of the data matrices, but also investigates the similarities and differences between them. In addition, data structure is interpreted by investigating all of the relationships, defined as the columns or rows in the data matrix. It is a wide

Related Search

Similar documents

We Need Your Support

Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks