Refining a socio-economic status scale for use in community-based health research in IndiaP Dudeja, P Bahuguna, A Singh, N Bhatnagar
School of Public Health, Post Graduate Institute of Medical Education and Research, Chandigarh, Punjab, India
Correspondence Address: Source of Support: None, Conflict of Interest: None DOI: 10.4103/0022-3859.150442 Clinical trial registration Not Applicable
Source of Support: None, Conflict of Interest: None
Clinical trial registration Not Applicable
Objective: Socio economic status is an important determinant of health and disease in population. Various scales for measuring the same exist in modern Indian society each with it's own limitations. Present study was done to abridge the existing and latest available Aggarwal Scale. Study Design: Cross Sectional Study
Keywords: Scale, research, socio-economic, factor analysis, principal component analysis (PCA), aggarwal
Socio-Economic Status (SES) signifies the position that a person or his family enjoys in the society. It is an important social determinant with significant bearing on the health, nutritional status and morbidity and mortality patterns of people. SES is an important component of any community health-related research. It also plays a significant role in planning and execution of development programs. There have been many scales for different settings with their own limitations. Hence, there is a need for development of a valid and reliable tool for measurement of SES. 
Attempts have been made to classify different sections of society according to their SES.  Different scales have been developed for use in community-based research in India. Till 1960, health-related research studies in India used occupation-based classification advocated by the British Registrar General.  In many western countries, occupation was the basis of social stratification. Scales like Hollingshed (Education and Occupation), Duncan (occupation), Nam and Powers (Education, Income and Occupation) and National Statistics Socio-economic classification (Occupation) incorporated education, income and occupation in various combinations for assigning SES  Researchers in India have attempted to classify SES using multiple parameters. Prasad's classification based on per capita monthly income evolved in 1961.  It was later modified in 1968 and 1970. , However, the major limitation was that often, incomplete details about income would be obtained. Presently, the Kuppuswamy scale (1976), which is based on education, occupation and income of the head of the household is used in urban areas. , Later, Pareekh (1981) evolved a classification based on nine characteristics for rural communities. 
Flexibility and robustness of these scales have often been questioned. Scales till date do not account for social mobility to a great extent. Social mobility is the movement of individuals, families, households within or between social strata in a society. It is a change in social status relative to others' social location within a given society. 
To overcome these disadvantages, Aggarwal et al. devised a new scale which measures the SES of families in both urban and rural areas. Focus was shifted from the head of the household to the highest achiever in the family and accounted for income from all sources. The scale is comprehensive and includes 22 items.  Inclusion of a large number of parameters gave the tool accuracy and complexity, making it a time consuming and labor intensive exercise.  Hence, there is a need for developing an abridged version of the scale with selected key indicators from Aggarwal et al.'s questionnaire that might increase the precision and reduce the time needed for assessment of SES.
In view of the above, the present study was conducted to identify a relatively smaller number of factors (domains) from the present 22-item Aggarwal et al scale, for accurate estimation of SES through a data reduction technique.
The present study is a secondary data analysis of a research project conducted by the department of Community Medicine, PGIMER Chandigarh, India on establishing the reliability and criterion validity of socio-economic scales in the urban area in 2012.
This study was approved by the Institutional Ethical Committee and written informed consent was taken from all participants.
In the city of Chandigarh, the population distribution in urban, urban slum and rural areas is 60%, 30% and 10%, respectively. The average per capita monthly income in Chandigarh is Rs. 100,876/- and literacy rate is 86.4%. 
Randomly one urban sector and urban slum colony of Chandigarh was selected using random number table. A cross-sectional survey of these selected households was attempted using systematic random sampling technique. For any multivariable analysis like principal component analysis (PCA) minimum of 8-10 subjects are required for every variable. Thus, for a 22-item Agarwal et al. scale taking into consideration the non-response, a sample size of 200 was selected. Eventually, analysis of 197 households was performed as three forms had incomplete data.
Aggarwal et al.'s scale was used as the gold standard in this study and four abridged versions of the same were developed. First, a list of variables which were required to classify a household as per Aggarwal et al.'s scale was prepared. Detailed information was thereby obtained from the head of the household by interviewer (Master in Public Health students). Data recorded during the interview was used to determine the SES of each household using the Aggarwal et al. scale.
Data was analyzed using SPSS version 16. Initially, exploratory and later confirmatory analysis was done. Kaiser-Meyer-Olkin (KMO) index was applied to Measure Sampling Adequacy (MSA). , Bartlett's test of spherecity was applied to test the null hypothesis that there is no correlation between the variables of the scale. (This hypothesis stated is for PCA tests and is not applicable for the entire study.)
The data reduction technique used in the study was Factor Analysis (FA) under which PCA was done. ,,,
Prior to performing PCA, the suitability of data for factor analysis was assessed by generating a correlation matrix. A correlation matrix of all the 22 variables in Aggarwal et al.'s scale was prepared. The matrix was inspected for coefficients with values ≥0.3. Correlated variables were extracted from the matrix and Eigen values were calculated. Components were selected using the following criteria: Eigen values of more than one and a break in the scree plot.  Thus, the 22-item scale was divided into four major components: Prominence, paying capacity, assets and affordability. Using EFA, the Agarwal et al. scale was abridged considering only the items with loadings above 0.6. Varimax rotation was performed to abridge the scale, using only the items with strong factor loadings more than or equal to 0.6.
Classification of categories using thresholds for reduced scales was tested between 15% and 20% intervals to arrive at optimal solution. Optimization was done looking at the maximum percentage of agreement between reduced scales and Agarwal et al. scale as gold standard. Thresholds were set based on the percentage score of an individual out of maximum score, that is, different for different abridged scales with varying number of items. For six and nine item scale, optimization of agreement was done at intervals of 18% threshold, that is, up to 18%, 18-36%, 36-54%, 54-72%, 72-90% and 90-100% representing very poor, poor, lower middle, upper middle, high and upper high socio-economic class, respectively, whereas for 11-item scale, optimization was done at 15% intervals.
The internal consistency of items in the different scales was assessed by calculating Cronbach's alpha. To assess the reliability, Intra Class Correlation (ICC) coefficients were also calculated for two different gradings of the same households using Aggarwal et al.'s scale and the reduced scales.
All the 197 households were rated using Agarwal et al.'s scale and their SESs were calculated. The distribution of households as per occupation, income and education is given in [Table 1].
Kaiser-Meyer-Olkin index and Bartlett tests
The KMO value was found to be 0.89 for our study sample against the recommended value of 0.6, which indicated that sampling was adequate and factor analysis was permissible. Bartlett's test was significant; supporting the factorability of the correlation matrix.
The correlation matrix showed positive correlation between family possessions, occupation and locality and monthly per capita income [Correlation coefficient (r) = 0.615, 0.516 and 0.677, respectively]. Other correlated variables were occupation and locality, locality and family possessions and domestic servants and social status (0.678).
Principal Component Analysis (PCA)
PCA revealed the presence of 6 principal components with Eigen values above 1 but the scree plot elbowed at the 5th point suggesting that the first four components should be selected [Figure 1]. These 4 components explained 21.2%, 12.6%, 8.7% and 6.6% of the variance respectively. All the 22 items of Agarwal et al.'s scale were divided among these four components called prominence, paying capacity, assets (parental support and land for cultivation) and affordability based on their factor loadings [Table 2].
Items under the first component (prominence in society) with strong factor loading were locality, education of husband/ wife, occupation of husband/ wife, family possessions, caste and monthly per capita income.
Under component 2 (paying capacity), loadings above cut-off criteria were shown by possession of other house, vehicle and income tax paid.
Parental support and land for cultivation loaded dominantly under the 3 rd component (assets).
The number of children and family members with international travel experience loaded significantly under the 4th component (affordability). This resulted in the formation of 4 Reduced Scales (RS) - RS-1 with 6 items, RS-2 with 9 items (containing items from components 1 and 2), RS-3 with 11 items (containing items from components 1, 2 and 3) and RS-4 with 13 items (containing items from all 4 components).
As the study sample was collected from Chandigarh which is predominantly an urban population, two items of component 3 (family support) failed to determine the SES of households.
The ICC coefficients between Agarwal et al.'s scale and RS-1 to 4 were estimated to be 0.786 (0.716, 0.838), 0.915 (0.888, 0.936), 0.92 (0.894, 0.94) and 0.952 (0.937, 0.964), respectively [Table 3].
Agreement was ascertained by grading 197 households using Agarwal et al. scale as the gold standard against different reduced scales. Percentage reduction in number of items from Agarwal et al.'s scale to the reduced scales 1-4 are 73%, 59%, 50% and 40%, respectively. Agreement between Agarwal et al.'s scale and reduced scales 1 - 4 was 40%, 73%, 73% and 85%, respectively. Scales 2-4 showed good agreement (>/= 70%) with Agarwal et al.'s scale [Table 4]. Items included in the abridged scales and their scoring pattern for calculation of socio-economic status has been shown in [Table 5] and [Table 6] respectively.
As per the definition of modernization, "ascribed status" gives way to "achieved status" as it legitimizes social gradation.  Ascribed status is a position based on who a person is and not what he does. Achieved status is a position which a person gains based on merit. Traditional Indian society was categorized on the basis of caste. In this system there was no possibility of an individual moving up or going down in the social hierarchy. People continued to remain in the group in which they were born. 
In the modern era, status immobility surrenders to rapid vertical and horizontal mobility.  Achieved status of children is growing with rising education which is better than that of their parents. This leads to better jobs and an upward shift in their socioeconomic statuses.  However, the available SES scales (Rahudkar scale 1960,  Udai Parikh scale 1964,  Jalota Scale 1970,  Kulshrestha scale 1972,  Kuppuswamy scale 1976, Shrivastava scale 1978,  and Bharadwaj scale 2001.  ) are not sensitive enough to capture sudden changes in status that can result from changing occupation of son, sale of land or sudden increase in property worth.
The Wealth index used in Demographic and Health Surveys (DHS) is a purely economic index and hence, cannot be used as a proxy measure for SES. The National Family Health Survey has used the Standard of Living Index to classify households into different socio economic divisions but ignored education and occupation as variables for assessing the same. Other scales developed by Gupta and Mahajan or Prasad etc have relied on income as the sole measure for assessment of SES of the family. , This often leads to error since income is not correctly revealed by Indian families.
Among all of these options, the most widely accepted scale for urban populations has been the Kuppuswamy scale. It is concise but over dependent on parameters of the head of family. Along with this limitation there remains a huge demand for updated versions because of changes in the inflation rate. ,
The 22-item long Agarwal scale has tried to overcome this problem. The scale is comprehensive with wide utility across urban and rural areas. It takes into consideration both direct and indirect measures for assessment of the financial status of the family. It gives due weightage to SES through several other family parameters, namely caste, material possessions (vehicle, TV, fridge, AC, washing machine, mobile phone, house, tap water, domestic servant, land, milch cattle etc), visits abroad, monthly per capita income, income tax paid, highest education in the family and occupation.
However, the length of the Agarwal questionnaire makes it a time and labor-intensive exercise. There are many other shortcomings to this scale. Several items are not relevant to urban areas; calculation of pooled income of the family is difficult as it is unclear whether to include income of the individual who is earning but not contributing to the family purse; occupation scoring is also difficult as many categories are not mentioned, for example, lower rank government servants; it is unclear how to assess land possession (agricultural, nonagricultural) and inheritance of property. This can lead to miss-classification by data collectors.
The present analysis narrowed down the 22 items from Agarwal et al.'s scale to six factors in RS-1, namely, locality, education of husband /wife, occupation of husband / wife, family possessions, caste and monthly per capita income. These variables were selected as they had Eigen values of more than 1. The same was confirmed by construction of a scree plot. These 6 factors account for 49% of the variation and can be taken as a surrogate measure of SES of the family. It is to be noted here that income, education and occupation are included in this list of 6 factors. These are also utilized by commonly used scales for SES calculations, namely Kuppuswamy's and Agarwal et al.'s scales.
There were nine factors in RS-2, 11 in RS-3 and 13 in RS-4. Reduced scale 2 had components of:
The first reduced scale has an agreement of 40% with Aggarwal et al.'s scale when tested on study subjects. The percentage agreement increased to 73% when we increased the items to 9 and then to 11. Maximum agreement of 85% is obtained with inclusion of 13 items from Agarwal et al.'s scale.
Percentage agreement of the reduced scales with Agarwal et al.'s scale was satisfactory. Cronbach's alpha values of reduced scales were high suggesting a good internal consistency. Interclass correlation coefficient values also supported the reliability of reduced scales. However, further research is warranted to test the reduced scales evolved in our study against the Kuppuswamy and Prasad scales.
The decision to select a scale for research depends on making a trade-off between investigator burden, resources, time availability and score precision. Abridged versions have varied utility in day to day life. In the case of studies where socioeconomic variables need to be recorded as demographic variables, the smallest version with 6 items can be used. Similarly if SES needs to be documented as an independent variable or has a significant bearing on the research question and results, abridged version with 9/11 items can be used [Table 5].
Scale development is an ongoing process and our aim in the present study was to shorten the scale to facilitate its easy usage by researchers. In other fields also, it has been observed that the research fraternity strives to develop abridged versions of comprehensive scales. WHO Quality of Life questionnaire SF 36 had 36 variables and placed considerable burden on both the respondents and investigators.  This led to the development of SF-12 with 12 items, requiring less than a third of usual time needed to complete SF-36.  On similar lines, many scales have two versions for practical usage e.g. Sense of Coherence (SOC) scale (SOC-29, SOC-13), Emotional Processing Scale (EPS-38, EPS-25), Berger Scale for assessing stigma of HIV/AIDS cases, International Index of Erectile Function (IIEF- 15, IIEF-5) etc. ,,
The strength of the present study is that we followed an iterative process involving factor analysis approach and theoretically driven statistical methods for abridgement of the existing scale. In order to respect the original scale we did not alter the response option or class intervals. Limitations of the study were that abridged scales were not administered to separate study groups and time required for the same was not measured. However, we did attempt to find the percentage agreement between the Agarwal et al. scale and reduced scales. Despite this limitation, we hypothesize that validated and abridged scales have relevance in a variety of settings. We propose that these versions can be used in different studies based on the research question, sample size, time available for data collection etc.
[Table 1], [Table 2], [Table 3], [Table 4], [Table 5], [Table 6]