We examined the income situation of the population in the South Transdanubia region by several regression methods at settlement level: log-level (OLS) model, ordinal generalized linear model (OGLM) and spatial error model were used. We have identified those principal components formed from variables which have the greatest effect on the total domestic income per capita in the region, and which determine the median income band of the population of a given settlement. The economic environment and the aging have the greatest impact on the income situation among the five principal components developed in the study. The spatial error model, which was applied due to the presence of spatial dependence, using Local Moran I, also confirms the existence of spatial autocorrelation in the income situation of the population of the settlements: the value of the spatial autoregressive coefficient is positive and significant. The results of the regression methods show that the spatial error model explains the income situation of the population of the 656 settlements more accuretely than the log-level (OLS) model.
The Southern Transdanubian region consists of three counties: Somogy, Tolna and Baranya. It covers an area of 14.169 km^{2}, where according to data of HCSO (Hungarian Central Statistical Office) 894 223 people lived on 1st January 2017. There are 615 municipalities and 41 cities in the region. In accordance with the national trend, the population is constantly decreasing by thousands by year, on 1st January 2014 still 917 492 people lived in Southern Transdanubia, and in 2010 the population was 947 986. The examined region is the least populated region of the country, with a population density of 63/km^{2} in 2017, compared to the national average of 105/km^{2}. (Somogy: 51/km^{2}, Tolna: 60/km^{2}, Baranya 83/km^{2}) (HCSO, 2019). The age distribution of inhabitants shows a similarly unfavourable trend, the population of the region is constantly aging, the proportion of children is decreasing and the proportion of elderly is increasing. In 2017, the region had the third oldest population in the country, only the Southern Great Plain and Budapest had more unfavourable data. More than fifty percent of the settlements in the region are small villages (population under 500 inhabatitants), and there are lot of dead-end villages. These factors cause significant economic and transportation disadvantages (Arany, 2005).
The amount of gross domestic product (GDP) in 2017 in Southern Transdanubia was HUF 2312.4 billion, which represented 6.02% of the national GDP. GDP per capita in 2017 was HUF 2.597 million, which is 66.3% of the national average (HUF 3.919 million). Southern Transdanubia had the 6th position in the GDP per capita ranking among the 7 hungarian NUTS-2 statistical regions, surpassing only the Northern Great Plain region (HCSO, 2019).
The unemployment rate in the region was 6.3% in 2017, this is a significant difference from the national average (4.2%). (HCSO, 2019). The employment rate of the 15-64 age group was 63.0%, which was lower than the national average (68.2%). The average monthly gross earnings of full-time employees in Southern Transdanubia in 2017 were HUF 244 203, less than the national average (HUF 289 671).
In our research, examining the settlement-level income data of the South Transdanubia region, we try to reveal which factors affect the income situation of the population.
More and more researchers point out that GDP is not necessarily a good measure of the overall socio-economic situation, either at the macro level or at the regional level. Therefore, in the studies of recent years, in addition to the various human development indexes (Horváthné Kovács et al., 2017), income has also become one of the central indicators of development examination.
The income situation and the differences in it often also mark the center-periphery boundary line. The results of research on center-periphery dichotomy connect income levels closely to other important spatial indicators, such as employment or commuting which actually mark the boundaries of the agglomeration of cities (Pénzes, 2013).
If we compare the studies examining the income situation of the settlements of the South Transdanubia region (Csizmadia, 2017) e.g., with the works researching the complex development of the settlements in the region, we conclude that incomes and complex development are affected by roughly the same indicators (spatial, economic, infrastructural, social, employment indicators), in the same direction.
The population of a given spatial unit is an important and recurring indicator in researches using regression analyzes, either at the micro-regional (Kiss–Németh, 2006) or agglomeration (Lôcsei, 2004) level, as a factor affecting income and income inequality. According to some results, the relationship between population and income level is close (Molnár – Ilk, 2010), but according to other researches, this close relationship between settlement size and income level cannot be detected in all cases (Lôcsei, 2004).
During her research, Dóra Szendi (Szendi, 2017) examined the differenes of territorial GDP and income and researched the role of the spatial dimension. Furthermore, which is even more closely related to our research, Szendi analyzed the distribution of income per full-time employee with the application of principal component analysis and regression models for 610 settlements in the Northern Hungary region, and also aimed to determine the factors influencing income. Neighborhood effects were tested by spatial lag and spatial error models. Presence of business organisations, living conditions or comfort, demographic processes (natural reproduction, migration) and tourism play a significant role in the income distribution of the settlements in the Northern Hungarian region. As a result of demographic processes, the value of income decreases in the settlements of the region, the other principal components cause a positive shift (Szendi, 2017 p. 168-169.).
In the international literature, the study of income inequality is closely related to the so-called convergence debate, where the basic question is whether economic growth increase or decrease income inequalities in the world and in some of its regions (Major, 2001) (Kiss – Németh, 2006). In international studies, income impact assessments usually use a methodological procedure that also takes into account geographical factors and/or neighborhood effects, as well as spatial autocorrelation: geographically weighted regression (GWR), spatial lag or spatial error model. The results show that these methods are better able to model the distribution of per capita income than the traditional regression models (Chasco et al., 2008), (Yildirim et al., 2009), (Kalogirou – Hatzichristos, 2007). The most commonly used explanatory variables in this research are: population of settlements, education, age, houshold size, presence of business organisations.
However, examinations of income processes also have difficulties. Based on data between 1998 and 2002, Dusek (2004) proved at different spatial levels that gross wages are affected by territorial differences (territorial units, settlement type, settlement size, geographical location, neighborhood effect). He noted that “income disparities at a given time do not fully reflect differences in economic well-being due to territorially different price level and non-market activities, including the sometimes significant small-scale agricultural production. (…) The economic well-being is a narrower category than total well being as it includes factors whose magnitude can only be estimated” (Dusek, 2004 p. 1.). The limit of income researches primarily caused by the following situation: due to tax rules, only about 40-60% of income from legal activities appears in tax base of personal income (Pénzes, 2011). Despite of this, income is the most suitable variable for examining well-being, as it accounts for the largest share of money which population received, and we have the most detailed sources about it.
We obtained the required data for our examination from the database of the Land Information System (hereinafter: LIS). The data managed by LIS come from the Hungarian Central Statistical Office (HCSO), the National Tax and Customs Administration of Hungary (NTCA), local municipality, state and county offices. We used data from 2017, as this is the last year in the LIS database for which all the variables we considered important were fully available at the settlement-level during the research period. The only exception are education data, as settlement-level data of education are only available from the year of the last census, 2011. Table 1 shows the variables and their calculation methods. We examined the data of 656 settlements in our research.
19 explanatory variables were included in the study, but their amount would results in too complicated regression models, so we reduced the number of variables with principal component analysis (Sajtos – Mitev, 2007). With this method the original amount of information can be reproduced with significantly fewer variables for easier interpretation. Furthermore, principal component analysis can be used to filter out multicollinearity between the variables, the created principal components do not correlate with each other at all.
The correlation between the income situation of the population of the settlements and the created principal components was first analyzed by OLS (log-level) regression. In this case, the dependent variable of the model was the total domestic income per capita (i.e. gross average income).
The average income does not always give an accurate picture of the income situation of the population. An examination of median incomes would be useful, but data are not available. Therefore, we determined median income bands (Csizmadia, 2017) for each settlement.
It would be advisable to use an ordinal logistic regression analysis (ologit) on the median income bands, since in this case the result variable of our regression function is a categorical variable, the median income bands of settlements fall into categories between 1 and 9, where the higher category means a higher income situation. But ordinal logistic regression violates the parallel regression assumption in most cases, based on the Brant test, the model is not applicable. The method of OGLM was born for this problem in Stata software, this process treats the violation of the assumption of the ologit (Williams, 2016). OGLM estimates Ordinal Generalized Linear Models.
The OGLM, like the ordinal logistic regression model, is an extension of the logistic regression model. OGLM allows more than two (ordered) response categories instead of dichotomous dependent variables. In this model, the interpretation of the beta coefficients is as follows: that if the explanatory variable is increased by one unit, the dependent variable changes with the value of the given beta coefficient on the ordered scale. (McCullagh, 1980).
Studies that explore factors influencing the developmental conditions or income situation already make extensive use of regression models that also takes into account spatial effect, spatial dependence, for example spatial lag, spatial error model, or geographically weighted regression (GWR) (Szendi, 2017) (Chasco, et al., 2008) (Kalogirou – Hatzichristos, 2007).
The two most common methods for econometric modeling of spatial autocorrelation are the spatial lag and the spatial error model (Varga, 2002). These models are suitable for filtering out spatial effects. The variables that are still significant after filtering out spatial effects are the really important variables, these explanatory variables need to be influenced to increase the value of the dependent variable. In our study, we used the spatial error model.
The spatial error model is a method based on the correction of spatial autocorrelation between error terms. It clears the regression equation from the spatial autocorrelatoin effect of explanatory variables and the independent variable.
The general formula of the spatial error model is as follows:
Y_{(}_{N*1}_{)} = Xβ_{(}_{K*1}_{)} + ε, and
ε_{(}_{K*1}_{)} = λWε + ξ, where
Y: dependent varaible,
X: explanatory (independent) variables,
β: regression coefficients,
ε: vector of error terms,
W: weight matrix,
λ: autoregressiv coefficient (spatial error coefficient),
ξ: vector of uncorrelated error terms (Varga, 2002).
In the spatial error model, λ (lambda) indicates the spatial autocorrelation between the error terms, if its value is not 0, models that do not take into account spatial effects (e. g. OLS) will result in inaccurate estimates (Tsionas, 2019).
4.1. Examining the adequacy of variables for principal component analysis
Our sample size corresponds to the sample size prescribed in the literature, as we examined the data of 656 settlements by principal component analysis, which is well above the minimum sample size. The proportion of variables and settlements also meets the requirements, as the number of cases (in this case number of settlements) is more than 10 times higher than the 19 variables.
The 19 explanatory variables (X_{1}-X_{19}) in Table 1 were included in the principal component analysis. Checking the adequacy of the data for principal component analysis consists of several steps. First of all, it is important that there be a significant relationship (correlation) between the variables, moreover, multicollinearity is a necessary criterion, because without it, “similar” variables could not be found and compressed into a principal component (Csallner, 2015). The fulfillment of this criterion can be examined using a correlation matrix between the variables. According to our results more than 75% of the total correlation values shows multicollinearity significantly, so based on the correlation values, it can be concluded that our data are adequate for principal component analysis.
The next step in examining the adequacy of the data is to prepare the anti-image covariance and anti-image correlation matrix. The off-diagonal elements of the anti-image covariance matrix show the part of the variance that is independent of the other variables. It is a generally accepted rule that no more than a quarter of the elements outside the main diagonal may be greater than 0.09 (in absolute value.) If this criterion is not met, there is no close relationship between the variances of the variables, so we can not assume underlying relationship between the variables (Sajtos – Mitev, 2007). In our case, only 14 of the 171 values are greater than 0.09, so it can be stated that our data are also adequate for performing principal component analysis in this respect.
In the anti-image correlation matrix, the MSA (measure of sampling adequacy) values in the main diagonal range from 0 to 1. This value shows how closely the variable is related to all other variables. A value of 1 means that the given variable is estimated by the other variables without error. Variables with MSA values below 0.5 should be excluded from the analysis (Cerny – Kaiser, 1977). Outside the main diagonal, partial correlation coefficients are shown, the low values of coefficients indicate strong background variables/principal components. In the anti-image correlation matrix of the variables used in our study, there are only 1 in the main diagonal, and the values outside the main diagonal vary between 0.0023 and 0.7240 (but the latter is an outlier). Consequently, we can say that the elements outside the main diagonal take on a rather low value, so our data also meet this assumption of principal component analysis.
The 19 variables are also adequate for principal component analysis based on the Bartlett test and the Kaiser-Meyer-Olkin (KMO) criterion. According to the null hypothesis of the Bartlett test, the variables are uncorrelated in the population. Since in our case the significance level is less than 0.05, we reject the null hypothesis, i.e. there is a correlation between our variables, so they are adequate for principal component analysis (Table 2).
Table 2: Bartlett test and KMO criterion
Bartlett test Chi-square = 6690.063 Degree of freedom = 171 Sig. (p-value) = 0.000 |
H_{0}: the variables are not correlated |
Kaiser-Meyer-Olkin criterion KMO = 0.888 |
Source: Own construction based on STATA
The principal component analysis can also be performed with the examined variables on the basis of the KMO criterion (Table 2), because if the KMO value is greater than 0.8, it is considered a very good result. (Kaiser, 1974). The KMO value is the average of the MSA values. „While the MSA value applies to single variable, the KMO applies to all variables simultaneously” (Csallner, 2015).
4.2. Determination of the number of principal components
We chose from the many methods the Kaiser criterion (Eigenvalue), the cumulative percentage of total variance, the Scree-test, and the Akaike (AIC) and Bayesian information criterion (BIC) to determine the number of principal components (Table 3).
Finally, after Varimax rotation of the principal components (Table 4), we decided to create 5 principal components. This structure contains 67% of the original amount of information.
Table 3: Number of principal components according to different criteria
Method |
Number |
Kaiser criterion (Eigenvalue) |
5 |
Cumulative percentage of total variance |
4 |
Scree-plot |
5-7 |
AIC |
10 |
BIC |
7 |
Forrás: own constuction
4.3. Specification of the regression models
The principal components presented in Section 4.2 were used in the regression models, in all cases the dependent variable was the total domestic income per capita, the principal components were the independent variables.
According to our assumption, the principal component of economic environment will have a positive effect on the income situation of the population, as the higher presence of business organizations and entrepreneurs has a beneficial effect on local job creation, thus increasing the income level or reducing unemployment rate. So it is in connection with another variable of the principal component.
The unemployment rate has a negative sign in the factor, so the low unemployment rate results in a higher value in the average income of the inhabitants. Higher education attainment is an advantage for filling higher-ranking and better-paid jobs, which also points in the direction that the sign of this principal component will be positive in the regression analysis. Population density presupposes a greater presence of the service sector, which may also promote income growth. Coverage of internet subscriptions and the development of the gas network provide such infrastructural advantages for each settlement, which enables businesses and wealthier individuals to settle there, thus resulting in an increase in the income level.
Related researches support (Chasco et al., 2008; Szendi, 2017) our hypothesis: higher presence of business organozations, higher levels of education, lower unemployment and higher worker activity result in higher incomes.
The principal componenet of public utility services is also expected to have a positive sign in the regression model. The higher public utility services presupposes that real estate prices are also higher, so it is expected that people with higher incomes will buy real estate in these settlements. In particular, the coverage of a public sewerage network can be a critical consideration for people with higher incomes when choosing residential property, as other public utilities included in the principal component (waste transport, water and electricity supply) are nowdays commonplace.
In the literature, the favorable development of similar variables has a positive effect on the income situation (Szendi, 2017).
The princiapl compoenent of aging factor will certainly have a negative sign in the equation of the regression model, because if the proportion of children and the number of live births are low it will lead to the aging of the settlement, which reduces the proportion of active workers, thus negatively affecting the income level of the population. The high proportion of older people can also have the same effect on income. However, the results of similar research in the literature show that a higher number of children in the family reduces the average income. (Kalogirou – Hatzichristos,2007).
Assessing the impact of the principal factor of small-scale farmers factor on the income situation is difficult to decide. It is expected that it will have a negative effect if we think in terms of the following: as the annual income of a small-scale farmer can be maximum HUF 8 million, this means that their average monthly income can not exceed HUF 666,666 per capita. If small-scale farmers achieve only a portion of this income and deduct their costs from it, it results in lower-than-average level of income. Based on this, if the proportion of small-scale farmers in a settlement is higher, it will have a negative effect on the income situation. On the other hand, if we consider the small-scale form of agricultural enterprise as a supplementary activity, it can even result in an increase in income if there are as many small-scale farmers as possible in a settlement (Varga – Sipiczki, 2017; Sipiczki – Rajczi 2018).
We expected that the principal component of internal migration would have an increasing effect on the income level of settlements. Immigrants are likely to move to a given settlement for a job opportunity, thus increasing the proportion of workers there, which has a positive effect on the average income of the population. Furthermore, young people and graduates are more willing to emigrate from their place of residence, and if we assume that graduates work for higher wages, then immigrants to a given settlement may increase the income level of the population of that settlement. According to a 2011 survey, 70.9 percent of young people changed counties from their original place of residence because of their studies in higher education or post-graduate working (Nyüsti–Ceglédi 2013).
4.4. Results of regression analysis – log-level regression model
To eliminate heteroskedasticity, we calculated the natural logarithm of the original dependent variable (total domestic income per capita), and this value became the dependent variable of the regression equation during the analysis (log-level regression model). We ran a robust regression model because this model is less sensitive to measurement errors in the database or to the fulfillment of conditions recorded in the models (Sándorné Kriszt et al., 1997). Thus, it results in a more reliable, stable estimate, but we also performed the tests with a simple OLS model to check the heteroskedasticity. The resulting regression function satisfies the relevant conditions, including the conditions for heteroskedasticity and multicollinearity.
Analyzing the results (Table 5), we can state that the variables in the model explain the amount of total domestic income per capita in 57.9 percent. According to the model, the first four principal components have a significant effect on incomes in the settlements of the region. We can interpret our regression function as follows: if all factors are 0, then the total domestic income per capita in the region is HUF e^{13.54485} Ft, that is HUF 762 875. Increasing the principal component of the economic environment has a positive effect on income, if the value of the principal component increases by 1 point, then the total domestic income per capita will increase by
100 * (e^{0,1978008} – 1) percent, that is 21.87%. The interpretation of the β-coefficient in the log-level model is as follows:
%ΔY = 100 * (e^{β} – 1). Hereinafter, only the calculated percentages are reported. Increasing the value of the principal component of public utility services also improves the income situation, increasing it by 1 unit, we achieve a 1.76% increase in income. While raising the points by 1 unit of the principal component of aging and of small-scale farmers will reduce the total domestic income per capita. The former cause a decrease of 7.21%, the latter results in a decrease of 2.81%. The internal migration is not a significant variable according to the model.
The standardized beta coefficient in the last column of the table measures the strength of the effect of the explanatory variables on the outcome variable. The higher the absolute value of the beta coefficient is, the greater the effect is, so the effects of the explanatory variables become easily comparable (Freedman, 2009). Based on this, we see that the economic environment has the greatest impact on the total domestic income per capita in the region, followed by the aging, the presence of small-scale farmers, and finally the public utility services has the lowest efficiency.
4.5. Applying OGLM to median income bands
„The median income band is that income band (determined by NTCA), which among the income bands in ascending order cumulatively includes for the first time at least half of the taxpayers living in the settlement. Thus, at least half of the taxpayers living in the given settlement earn a maximum of gross median annual income belonging to the median income band (Csizmadia, 2017).” (See Appendix 1 for the income bands defined by NTCA.) The higher the median band of a settlement is, the more high-income residents live in that settlement.
The median income band is a categorical variable, it forms the dependent variable of the model, so an ordinal generalized linear model (OGLM) can be applied. The range of explanatory variables is still the 5 principal components which already are described above.
The pseudo-R^{2} of the applied ordinal generalized linear model is much smaller than the R^{2} of the log-level models. But it is important to note that the pseudo-R^{2} developed by McFadden and the R^{2} of the linear models are not fully comparable (McFadden, 1974). There is no fully accepted agreement on what pseudo-R^{2} is acceptable, but online professional forums on statistics interpret pseudo-R^{2} values between 0.2 and 0.4 as a good or excellent fitting in ordinal generalized linear models and logistic regression models.
Comparing the significant principal components and the signs of the beta coefficients, we see that the same principal components are significant in the log-level and ordinal generalized linear model (Tables 5 and 6). The only minor difference is in the case of public utility services: in the former model it is significant at the p<0.01 level, in the latter at the p<0.1 level. The signs of the beta coefficients of the significant principal components are exactly the same in the two models, which means that the changes of each principal components affect the total domestic income per capita in the same direction as the median income band.
4.6. Spatial autocorrelation – application of the spatial error regression model
Geographical location plays a role income situation of the inhabitants of the settlements, a spatial autocorrelation, dependence can be detected by examining the income situation. Income values have positive correlation with spatial location in the South Transdanubian region, especially in the case of county seats and their agglomerations (Csizmadia, 2017). According to Kolber et al., (2019) the enterprises of the more developed regions are more territorially concentrated, which also affects the income conditions of the settlements.
Figure 1 illustrates the spatial dependence of the income situation in the South Transdanubia region. It can be seen on the LISA cluster map which was created on the basis of the Local Moran I calculation (queen contiguity), that the spatial autocorrelation can be observed mainly in the agglomeration of the county seats.
Based on the above results (Figure 1), it is reasonable to include neighbourhood effects in our regression analysis. The neighbourhood matrix (queen contiguity) can be integrated into the OLS regression model, so the regression analysis also indicates the spatial autocorrelation using the Moran I-value. Whether the application of the spatial lag or spatial error model is expedient can be decided on the basis of Figure 2.
The first step in selecting the appropriate model is to run the Lagrange Multiplier (LM) Diagnostics: LM-error and LM-lag tests (Table 7). Moran’s I values can also be read in addition to tests.
Since the results show that the spatial lag and spatial error models, as well as both robust tests, are significant (this confirms the existence of spatial regression), based on the recommendation of the literature (Figure 2) (Anselin, 2005) (Szendi, 2017) the spatial error model should be run because it has higher value.
The spatial autoregressive coefficient (λ) has a value of 0.413 and it is significant (Table 8). Based on the z-value expressing the stability of the spatial parameter, it is the third most stable coefficient in the equation. If the value of all principal components is 0, then according to the spatial error model, the total domestic income per capita is e^{13.546}, equal to HUF 763 753 in the region. The economic environment, the aging and small-scale farmers are the significant principal components in the model. The sign of the coefficient of the economic environment is positive, so the increase of the points of the principal components by 1 unit, if the other principal components remain unchanged, increases the total domestic income per capita by 100 * (e^{0,201} – 1) percent, that is 22.26%. 1 unit increase in the principal component of aging reduces income by 6.12%, while an increase in the principal component points of small-scale farmers reduces it by 2.18%.
The likelihood ratio test which was run after the creation of spatial error model is significant at the p <0.01 level. This indicates that spatial effects are present in the model.
The OLS and OGLM models give very similar results, there is only one difference in the significance level of the significant principal components between the 5-5 principal components of the 2 models (public utility services). Thus, we can say that there is no big difference in examining the income situation of the inhabitants of the region with average incomes or median income bands. The signs of the coefficients of the principal components are the same in the models.
Local Moran I and the spatial error model prove the existence of spatial dependence in the income situation of the population between settlements: the value of the spatial autoregressive coefficient is positive and significant. The spatial error model also supports the results of the log-level (OLS) and the ordinal generalized linear model.
The results of the spatial error model better explains the income situation of the population of settlements than the log-level (OLS) model. The values of R^{2} and log likelihood are higher in the spatial error model, and the values of Akaike and Schwarz criteria are lower, so the spatial error model fits better than the log-level model.
The results of our research show that the principal component economic environment has a positive effect, the principal component of aging has a negative effect on the income situation of the population in the South Transdanubia. These two principal components have the greatest impact on the income situation. The signs of the principal components are the same in all three regression models (OLS, OGLM and spatial error), and except the public utility services, the significance levels are identical in the three different processes. The sign of the coefficient of principal component small-scale farmers is negative in all three models, while the sign of public utility services is positive, however, the latter principal component is not a significant variable according to the spatial error model. Internal migration has no significant effect on income situation.
The principal component economic environment includes the following variables: educational attainment of the population, the widest possible presence of business organisations, the lowest possible level of unemployment, coverage of internet and natural gas network. Among these factors, education, presence of business organisations and development of infrastructure (internet and gas network) have a positive effect on incomes, and high level of unemployment rate reduces the income level.
The principal component of aging includes the proportion of the non-working age population (children and the elderly), the number of live births and the number of three- or more-room apartments. The income situation can be most affected by increasing the proportion of the working age population, based on the birth of children. It is less and less common in today’s society that different generations live in the same house, because of this families are less likely to look for larger houses. The large, multi-room aparments are typical in the smaller settlements of the Southern Transdanubian region. This is also due to the fact that these settlements are aging, which worsens the income situation of the population and companies are reluctant to settle in such an environment.
The results of each regression model show that the presennce of business organizations, educational attainment of the population, and demographic indicators play a prominent role in the income situation of the population in the region. These are the target areas, intervention points, where it is absolutely necessary to take measures during the planning and execution of programs which work for the development of the region.
The results also show that due to the spatial autocorrelation, it is important for decision-makers to take into account that the development of a settlement can have an impact on the surrounding settlements, and vice versa, the economic and social detachment of a settlement can also have a negative effect on neighbouring settlements.
Anselin, L., 2005. Exploring Spatial Data with GeoDa: A Workbook. Urbana: University of Illinois, Urbana-Champaign. http://www.csiss.org/clearinghouse/GeoDa/geodaworkbook.pdf (downloaded: 2019. 09. 10.)
Arany O. V. (2005): Dél-dunántúli régió szociális és gyermekvédelmi helyzetelemzésének összefoglalója. Kapocs. 4.3.1-27. http://www.ncsszi.hu/download.php?file_id=1238 (downloaded: 2019. 09. 20.)
Cerny, B. A. – Kaiser, H. F. (1977): A Study Of A Measure Of Sampling Adequacy For Factor-Analytic Correlation Matrices. Multivariate Behavioral Research 12.43-47.
Chasco, C. – García, I. – Vicénés, J. (2008): Modeling spatial variations in household disposable income with Geographically Weighted Regression. Munich Personal RePEc Archive, https://mpra.ub.uni-muenchen.de/1682/1/MPRA_paper_1682.pdf (downloaded: 2019. 08. 15.)
Csallner, A. E., (2015): Bevezetés az SPSS statisztikai programcsomag használatába. Szegedi TudományegyetemJuhász Gyula Pedagógusképzô Kar. Szeged. http://www.inf.u-szeged.hu/~banhelyi/okt/SPSS_2020tavasz/csallner-spss-javitott.pdf (downloaded: 2019. 08. 30.)
Csizmadia, A. (2017): Települések jövedelmi és iskolázottsági helyzetének vizsgálata a Dél-dunántúli régióban. Kaposvári Egyetem Gazdaságtudományi Kar. Kaposvár
Dusek, T. (2004): Területi jövedelmi folyamatok Magyarországon. II. Magyar Földrajzi Konferencia, Szeged. CD kiadvány.
Freedman, D. A. (2009): Statistical models: Theory and practice. Cambridge University Press, Cambridge.
Horváthné Kovács B. – Barna R. – Titov A. – Nagy M. Z. (2017): A humán tôke indexe a Dél-dunántúli régióban. Közép-Európai Közlemények, 10.3.55-72.
Kaiser, H. F, (1974): An Index of Factor Simplicity. Psychometrika 39.1.31-36.
http://cda.psych.uiuc.edu/psychometrika_highly_cited_articles/kaiser_1974.pdf (downloaded: 2019. 10. 11.)
Kalogirou, S. – Hatzichristos, T. (2007): A Spatial Modelling Framework for Income Estimation. Spatial Economic Analysis. 2.3.297-316.
Kiss, J. P. – Németh, N. (2006): Fejlettség és egyenlôtlenségek – Magyarország megyéinek és kistérségeinek esete. Budapest Munkagazdaságtani Füzetek BWP – 2006/8, Magyar Tudományos Akadémia Közgazdaságtudományi Intézet, Budapest. https://www.econstor.eu/bitstream/10419/108430/1/bwp0608.pdf (downloaded: 2019. 09. 30.)
Kolber, K. – Parádi-Dolgos, A. – Pavlin, R. K. (2019): A gazdasági társaságok adóterhét befolyásoló tényezôk vizsgálata. Economica 10.3-4.1-7.
Lôcsei, H. (2004): A vidéki agglomerációk fejlôdési pályája. Mûhelytanulmányok 2004/24. Magyar Tudományos Akadémia Közgazdaságtudományi Intézet. Budapest.
https://www.econstor.eu/bitstream/10419/108093/1/MTDP0424.pdf (downloaded: 2019. 09. 30.)
Major, K. (2001): A nemzetközi jövedelemegyenlôtlenség dinamikája. Doktori (PhD) értekezés. Budapesti Corvinus Egyetem, Közgazdaságtani Doktori Iskola. Budapest.
http://phd.lib.uni-corvinus.hu/271/1/major_klara.pdf (downloaded: 2019. 09. 30.)
McCullagh, P. (1980): Regression Models for Ordinal Data. Journal of the Royal Statistical Society: Series B (Methodological), 42.2.109-127.
McFadden, D. (1974): Conditional logit analysis of qualitative choice behavior. In: P. Zarembka: Frontiers in Economics.Academic Press. 105-142.
Molnár T. – Ilk B. F. (2010): A települések fejlettségének alakulása a Dél-Dunántúli Régióban. Acta Scientiarum Socialium. 32.1.29-38.
Nyüsti, S. – Ceglédi, T. (2013): Vándorló diplomások, diplomáért vándorlók. In: Garai, O. – Veroszta, Z. (szerk.): Frissdiplomások 2011. Educatio Társadalmi Szolgáltató Nonprofit Kft. Budapest. p. 173-207.
Pénzes, J. (2011): A jövedelmi szempontból elmaradott települések területi átrendezôdése a rendszerváltozás után. Földrajzi Közlemények, 135.1.59-69.
https://www.foldrajzitarsasag.hu/downloads/foldrajzi_kozlemenyek_2011_135_evf_1_pp_059.pdf (downloaded: 2019. 09. 05.)
Pénzes, J. (2013): A foglalkoztatottság, az ingázás és a jövedelmi szint összefüggései Északkelet- és Északnyugat-Magyarországon. Területi Statisztika. 53.3.202-224.
Sajtos, L. – Mitev, A. (2007): SPSS kutatási és adatelemzési kézikönyv. Alinea Kiadó, Budapest.
Sipiczki, Z. – Rajczi, A. (2018): Magyarország mezôgazdaságának jövedelmezôsége az agrártámogatások tükrében 2006 és 2015 közötti idôszakban. Közép-Európai Közlemények. 11.3.84-92.
Sándorné Kriszt, É. – Varga, E. – Veitzné Kenyeres, E. – Korpás, A. – Csernyák, L. (1997): Általános statisztika II. Nemzeti Tankönyvkiadó Rt. Budapest.
Szendi D. (2017): Települési jövedelmek térbeli kölcsönhatása: felzárkózás vs. tartós különbségek?. In: Gelei A. /szerk./: Sokszínû gazdálkodástudomány. MTA IX. Osztály, Gazdálkodástudományi Bizottság, Budapest. 141-174.
Tsionas, M. (2019): Panel Data Econometrics – Empirical Applications. Academic Press. London
Varga, A. (2002): Térökonometria. Staitsztikai Szemle, 80.4.354-370.
Varga, J. – Sipiczki, Z. (2017): A hazai mezôgazdasági vállalkozások gazdálkodása, a belsô felhalmozás, a hitelezés és a támogatási csatorna tükrében 2005 és 2014 között. Közép-Európai Közlemények. 10.3.138-151.
Williams, R. (2016): Understanding and interpreting generalized ordered logit models. The Journal of Mathematical Sociology 40.1.7-20.
Yildirim, J. – Öcal, N. – Özyildirim, S. (2009): Income Inequality and Economic Convergence in Turkey: A Spatial Effect Analysis. International Regional Science Review, 32.2.221-254.
Appendix 1: Income bands defined by NTCA
• Income band 1: gross 0 – 300 000 HUF/year
• Income band 2: gross 300 001 – 600 000 HUF/year
• Income band 3: gross 600 001 – 800 000 HUF/year
• Income band 4: gross 800 001 – 1 000 000 HUF/year
• Income band 5: gross 1 000 001 – 1 200 000 HUF/year
• Income band 6: gross 1 200 001 – 1 500 000 HUF/year
• Income band 7: gross 1 500 001 – 1 800 000 HUF/year
• Income band 8: gross 1 800 001 – 2 000 000 HUF/year
• Income band 9: gross 2 000 001 – 2 500,000 HUF/year
• Income band 10: gross 2 500 001 – 3 000 000 HUF/year
• Income band11: gross 3 000 001 – 4 000 000 HUF/year
• Income band 12: gross 4 000 001 – 5 000 000 HUF/year
• Income band 13: gross 5 000,001 – 10 000 000 HUF/year
• Income band 14: gross 10 000 001 – 20 000 000 HUF/year
• Income band 15: above gross 20 000 000 HUF/year
Source: Own construction based on NTCA
Adrián Csizmadia, PhD student
Szent István University Kaposvár Campus Doctoral School of Management and Organizational Sciences
Dr. Tibor Bareith, assistant lecturer
Szent István University Kaposvár Campus Faculty of Economic Science, Institute of Finance and Accounting
The research was supported by EFOP-3.6.2-16-2017-00018 in Kaposvár University project.
@ WCTC LTD --- ISSN 2398-9491 | Established in 2009 | Economics & Working Capital