Methodology


The Scope and Steps in Developing the lndex

The City Development Index aims to compare cities objectively from a holistic perspective both at the international and national levels. In order for the methodology of the City Development Index to be strong and the data infrastructure to be healthy, a meticulous and intensive work process was carried out at every step in creating the index. The process of creating the City Development Index consists of the following steps: identifying the theoretical framework and key areas, data collection and indicator creation, determining the geographical scope, data processing, weighting and aggregation, classification and sequencing, and visualization and presentation.

When comparing the City Development Index internationally, a total of 48 indicators were constructed under 3 domains (social, economic, education/culture) and 12 dimensions. The dimensions of demographic dynamism , social welfare, health & safety, and environment have 18 indicators, and the dimensions of economic wealth, development, openness, and work life have 16 indicators. In the domain of education and culture, 14 indicators were produced under the dimensions of education, human capital, connectedness, and diversity and participation.

Figure 1. The Process for Creating the City Development Index

First, we identified 100 cities from each continent with a population of over 1 million for comparing the City Development Index internationally; those that involved the most indexes were examined within the scope of the study. Next, 50 cities were selected from each of these 100 cities, paying attention to geographical inclusion and diversity. In particular, we selected cities that are prominent both in the sub-regions and in terms of country, paying attention to balance the distribution of cities from among 5 continents (Africa, America, Asia, Europe, and Oceania) and 9 sub-regions (Africa, Western Europe, Eastern Europe, South America, North America, Middle East, South and Southeast Asia, Central and East Asia, and Oceania).

Macroform borders or the municipal borders of medium-sized cities/metropolitan areas or of the entire province of large cities have been taken as the basis while determining the city areas for the international comparison. A total of 55 cities were compared by including 5 cities from Turkey with the 50 other cities selected from around the world.

Map 1. Cities Around the World Included in the City Development Index.


Table 1. The Indicator Structure of the City Development Index


Table 2. Cities Included in the Index

Data Generation and Analysis

Because the main approach of the City Development Index involves the availability of statistical quantitative data, the data collection phase directly affects the indicator construction process. The standardization, winsorization, normalization, and imputation stages occur respectively after collecting the raw data; thus, the indicator scores required for weighting are formed by processing the raw data.

Through about 10 months of meticulous work, the project team obtained the national and international raw data within the scope of the City Development Index from the databases of international organizations such as the UN, OECD, ILO, World Bank, UNESCO, Eurostat as well as from statistical offices of countries and cities. The first stage standardized the raw data for all the cities by converting the data to the same units of measurement (e.g., percentages, quantities, per capita value).

In order to prevent outliers in each indicator from dominating the index results, the second stage determined indicators with a skewness greater than 2 and a kurtosis greater than 3.5 and winsorized the outliers in these indicators to the nearest maximum or minimum value.

The third stage normalized all indicators using the method of adjusting to the maximum value’ in order to ensure that the indicators with different measurement units are on the same page and to avoid over- or under-representation in the index score. Thus, values were produced in the range of 0-100 for each indicator. The advantage of adjusting to the maximum value is that the pre-normalized structure of the data remains intact after normalization. Other normalization methods partially disrupt the structure of the data, narrowing distributions with wide ranges and widening distributions with narrow ranges. After normalization, negative indicators were subtracted from 100, and the direction of the indicator was changed.

The fourth stage estimated missing data for each indicator using imputation techniques if no data was available for a city or if deficiencies occurred in the temporal data of a city. In case of data missing for certain years in a city, the value of the missing years was estimated according to the structure of the data using a projection method by averaging the previous and following years, or using the most recently published data. If no data could be found for a city specific to that indicator, the city data was obtained by taking a relative proportion from the region and country data. If country data was not available, the missing data were completed by taking the average value of the dimension included in that indicator as its value.

Weighting the lndicators

After completing the data processing phases, the indicators were weighted and combined with the dimensions. The criteria importance through intercriteria correlation (CRITIC) method was used to weight the indicators. With this method, weighting is done according to two basic criteria: the first is the standard deviation value, which shows the spread of the observed values for the variable, and the second is the correlation coefficients, which shows the intensity of the relationships among variables. Accordingly, the weight of a variable is directly proportional to the standard deviation of that variable and inversely proportional to the correlation value between it and the other variables. The final stage combined the indicators using the weighted arithmetic average method while calculating the index score. Thus, the weight of a dimension was obtained by adding the weights of the indicators making up that dimension, and the weight of a domain is obtained by adding the dimensions that make up that domain.

The final stage also calculated the dimension, domain, and indicator scores for each city between 2010-2019 and ranked the cities individually according to the calculated indicator, domain, and dimension scores. Thus, the cities at the top, bottom, and middle were identified with respect to their dimension, domain, and index scores. The cities and nine regions where these cities are located were evaluated with respect to their 10-year average scores. As a result of the rankings, the cities were additionally classified in different categories with respect to both the rise and fall over the years and to their general rank.

Table 3. The Direction the Indicators Contribute to the Index, and Weightings for the Indicators, Dimensions, and Domains