Impact of public health expenditure on malnutrition among Peruvians during the period 2010-2020: A panel data analysis

Methodological design

The methodological approach considers the traditional quantitative approach and correlational type of research, given that it analyzes the impact of public health spending on malnutrition among Peruvians, using data from the National Household Survey, the Central Reserve Bank of Peru, the National Institute of Statistics and Informatics and the Ministry of Economy and Finance from 2010-2020.

Procedure

The data comprises a total of 2,112 observations covering the 24 departments of Peru during the period 2010-2020. The dependent variable is malnutrition, measured by caloric deficit, while the independent variable is public health expenditure, incorporating control variables such as inflation, rural population, unemployment, education, exports, and gross domestic product. The analysis was conducted over the 2010-2020 period based on data availability, considering inclusion and exclusion criteria.

The inclusion criteria were Peruvian households experiencing a caloric deficit. We defined caloric deficit according to the methodology used by Herrera (2002), which establishes a caloric norm for Peru of 2,318 calories per capita per day. This caloric norm is derived from data from the National Household Survey conducted by the National Institute of Statistics and Informatics between the years 2010 and 2020.

While the exclusion criteria is peruvian households not experiencing a caloric deficit according to the methodology used by Herrera (2002) which defines the caloric norm for Peru as a total of 2,318 calories per capita per day, based on data from the National Household Survey by the National Institute of Statistics and Informatics for the years 2010 to 2020.

Sample

The sample size consists of a total of 2112 observations covering the 24 departments of Peru during the period 2010-2020, considering that for quantitative research according to Hernández & Mendoza (2018) the minimum sample size for causal or comparative analysis is 64 data, which implies that the sample is non-probabilistic, given the careful and controlled choice of data that have fed the variables previously in the problem statement.

The sampling method is non-probabilistic as it addresses the entirety of the data, totaling 2,112 observations, involving departmental analysis for the period 2010-2020.

The unit of analysis comprises each department, totaling 24 departments in Peru.

Document analysis was used as the technique for measuring the variables, and the observation guide was employed as the instrument. This tool is used to systematically and structurally collect data during an observation or study.

Data were obtained from the microdata database of the National Household Survey for the years 2010 to 2020. The data were extracted from module 01 and 02 (Household Member Characteristics), module 03 (Education), module 04 (Health), module 05 (Employment and Income), module 07 (Food and Beverage Expenditures), and module 08 (Charitable Institutions) published on the website of the National Institute of Statistics and Informatics. This annual survey allows for the tracking of indicators on living conditions and is conducted nationwide across the 24 departments of the country.

Additionally, data on the percentage variation of the price index by department for the period 2010-2020 and the gross value added by department at constant prices from 2007, compiled by the National Institute of Statistics and Informatics, were used and published in its economic statistics section.

For the collection of public health expenditure data, departmental data reflecting the amount spent in millions of soles were obtained from the Economic Transparency Portal of the Ministry of Economy and Finance for the period 2010-2020.

Finally, the database of the value of all goods and other market services exported to the rest of the world by department, measured in FOB value in millions of dollars, from the Central Reserve Bank of Peru for the period 2010-2020 was used. This data reflects the registration of sales abroad of goods and services by a resident company in Peru, including both traditional and non-traditional products.

Data analysis method

The estimation of malnutrition is based on the estimation of the caloric deficit and the basal metabolic rate using the methodology by Herrera (2002) which defines the caloric norm for Peru as a total of 2,318 calories per capita per day. This is based on the data from the National Institute of Statistics and Informatics, considering the average weight estimates by sex and age, classified for children under 10 years and those over 10 years by FAO (Swindale & Ohri, 2004).

For calculating the basal metabolic rate (BMR) for those over 10 years old, corrections were made based on estimates for Latin American countries (Bengoa et al., as cited in Herrera, 2002). Two options were considered: moderate activity for those over 10 years old in urban areas and intense activity for those over 10 years old in rural areas.

From the data obtained, which include time series and cross-sectional data, which allows the analysis of multiple observations over time for the same sample unit relevant to each department of Peru analyzed in the period 2010-2020, the econometric model called panel data is used, since it allows obtaining the largest amount of information and at the same time controlling the unobservable variability that is constant over time but may differ between subjects, reducing bias in the results (Wooldridge, 2010).

The panel data model helps to analyze changes over time and to identify causal relationships between variables that may vary between periods and reduce collinearity for better accuracy of estimates, allowing to increase the efficiency and power of estimates, by obtaining more robust and reliable results (Wooldridge, 2010).

On the other hand, the Generalized Least Squares (GLS) model is used since it becomes a fundamental tool when working with structural equations that present problems of heteroscedasticity and autocorrelation (Wooldridge, 2010).

By applying the GLS model, it is possible to obtain regression coefficients that more accurately reflect the relationship between variables, even when the data present variability and dependence in the errors. This is especially important in structural equations, where the complexity of the relationships between variables can make problems of heteroscedasticity and autocorrelation more common and, therefore, the use of GLS is essential to guarantee the validity and precision of the results obtained (Wooldridge, 2010).

The panel data model, based on Mushkin’s theory, will be used for the estimation, showing the relationship between malnutrition behavior explained by public health expenditure (Calva & Ruiz, 2021). The model utilizes two equations.

Through this model, the impact of public health expenditure on the malnutrition of Peruvians is quantified. Malnutrition, measured by caloric deficit, is the dependent variable, while public health expenditure is the independent variable. Control variables include inflation, rural population, unemployment, education, exports, and gross domestic product.

To ensure consistent estimations and an adequate analysis, variables such as public health expenditure, rural population, education, exports, and gross domestic product have been transformed into logarithms. Additionally, to estimate the regression with control variables, the structural equations of the variables under analysis are used (Wooldridge, 2010).

Equation 1 establishes the relationship between the dependent variable malnutrition and public health expenditure by department 𝑖=1,2, …, 24i=1,2, …, 24 in the year t=2010−2020 t=2010−2020.

On the other hand, the incorporation of control variables for the robustness of the model includes variables such as inflation, rural population, unemployment, education, exports, and gross domestic product, which are reflected in the following equation:

In Equation 2, structural equations will be used to evaluate the interdependencies, expressing the mentioned interrelationship in four equations:

Given the presence of heteroscedasticity and autocorrelation in the structural equations, confirmed by Breusch and Pagan (1979) tests for heteroscedasticity and Wooldridge (2010) tests for autocorrelation, a Generalised Least Squares (GLS) model was employed. The GLS estimator accounts for these violations of classical linear regression assumptions, providing more efficient and unbiased estimates. Additionally, a Hausman (1978) specification test was conducted to determine the appropriateness of fixed or random effects.

For the estimation of the relevant control variables for the study according to the objective of the research that seeks to quantify the impact of public health spending on malnutrition of Peruvians during the years 2010-2020, the theoretical review and availability of the data described above have been considered, taking as a basis the theory of (Mushkin, 1962) where it explains that the behavior of malnutrition is explained by public spending on health and the control variables considered by (Calva & Ruiz, 2021) that includes inflation, rural population, unemployment, education, exports and gross domestic product in order to reduce the bias of the estimates; applying said model with the data of Peru in the period 2010-2020.

Panel data models are based on several assumptions that allow estimates to be consistent, efficient and unbiased, as detailed below (Beltrán & Castro, 2010).

The variables used for the model estimation are detailed in Table 1.

Data processing

For the estimation of the undernutrition variable, the National Household Survey database for the period 2010-2020 was considered. Eviews 12 is used for the estimation of the panel data model, considering the control variables that are subjected to the proposed equations to analyze the interrelationship between the variables. This allows for both descriptive statistical treatment and the estimation of the proposed panel data model and the Generalized Least Squares (GLS) model. An academic license for the use of Eviews 12 software is available, registered under the name Lindon Vela Meléndez. The license details are as follows: Serial number: Q1208886 - D49010AF - 9D854485. The software can be downloaded from the following link: http://www.eviews.com/download/student12.

Autocorrelation detection was performed using the Lagrange multiplier tests of Breusch & Pagan (1979) and Wooldridge (2010). In addition, the Hausman (1978) test was used to choose between a fixed or random effects model.

We assessed the presence of autocorrelation in the model residuals using two statistical tests. First, the Breusch and Pagan (1979) Lagrange multiplier test was applied, which examines the null hypothesis of no autocorrelation against the alternative of autocorrelation up to a specified order. Second, the Wooldridge (2010) test, designed specifically to detect first-order autocorrelation in panel data, was used.

To determine the appropriate specification of the model, we perform a Hausman (1978) specification test. This test compares fixed and random effects estimators under the null hypothesis that individual specific effects are uncorrelated with the explanatory variables.

Ethical considerations

Data for this study were obtained from the National Household Survey (ENAHO) conducted by the National Institute of Statistics and Informatics (INEI). The ENAHO, as a publicly available and freely accessible dataset on the INEI website, adheres to rigorous ethical standards to protect the confidentiality of participants.

Furthermore, ENAHO data by INEI are anonymised before being made publicly available. This ensures that individual responses cannot be linked to specific participants.

Although we cannot provide specific details about the consent process for each individual participant in the ENAHO survey, INEI, as a reputable government institution, is committed to ethical research practices and to obtaining informed consent.

In the ethical use of public data obtained by the INEI in the ENAHO, we adhere to the principles of intellectual honesty, truthfulness, transparency, human integrity, respect for intellectual property, justice and responsibility, the study aims to comply with the “Code of Ethics in Research of the Universidad César Vallejo, version 01; by University Council Resolution N° 0340-2021-UCV”.

We have used the data for research purposes only, as permitted by INEI’s terms of use. We have not attempted to re-identify any individual, and our analysis does not include any personally identifiable information.

By using publicly available anonymised data and adhering to ethical principles in our research, we aim to minimise any potential ethical concerns while leveraging valuable information for the benefit of public health research.

Regression of malnutrition with control variables

Table 2 shows the results of the panel data estimation, where it is observed that public health expenditure, rural population, unemployment, and gross domestic product are statistically significant at the 5% level, considering Equation 2.

It is observed that for each 1% increase in public health expenditure, malnutrition decreases by 2.6%. Conversely, a 1% increase in the rural population leads to a 4.6% increase in malnutrition. Additionally, a 1% increase in the unemployment rate results in a 2.3% increase in malnutrition, while a 1% increase in gross domestic product leads to a 4.3% decrease in malnutrition. The variables with the most significant impact during the analysis period 2010-2020 are the rural population and the gross domestic product.

Regression of public health expenditure with malnutrition and unemployment

Table 3 shows the analysis of public health expenditure considering structural Equation 3. The results indicate that chronic malnutrition and unemployment are significant at the 5% confidence level. It explains that malnutrition has a negative effect on public health expenditure due to the limited importance given to public malnutrition policies, with public budgets being prioritized for other sectors.

It was observed that for each 1% increase in the malnutrition rate, public health expenditure decreases by 0.16%. Meanwhile, a 1% increase in the unemployment rate leads to a 0.40% increase in malnutrition.

Regression of rural population with education

Table 4 shows the analysis of the rural population considering structural Equation 4. The results indicate that education is significant at the 5% confidence level, explaining that education has a negative effect on the rural population. This is mainly due to the trend that as the human capital of the population increases, there is a migration from rural to urban areas, moving to central zones to improve their income levels.

It was observed that for each 1% increase in the average years of education, the malnutrition rate decreases by 1.46%.

Regression of unemployment with education, malnutrition, and inflation

Table 5 shows the analysis of unemployment considering structural Equation 5. The results indicate that chronic malnutrition, inflation, and education are significant at the 5% confidence level. It explains that malnutrition has a positive effect on unemployment, as do education and inflation. This shows that high levels of malnutrition lead to high unemployment rates, indicating that nutritional imbalance causes problems in performing physical or intellectual labor, affecting worker productivity.

Additionally, the positive relationship between education and unemployment is explained by the excess accumulation of human capital in Peru, which exceeds the available job opportunities due to limited positions. Regarding inflation, the positive relationship with unemployment contradicts the Phillips Curve theory (1958) which suggests an inverse relationship between inflation and unemployment. According to Campoverde et al. (2016) this positive relationship depends on the economic context of each country. In Peru, the loss of purchasing power and delayed investment due to price fluctuations result in a decline in business confidence, creating an uncertain climate that impacts higher unemployment rates.

It was observed that for each 1% increase in the malnutrition rate, unemployment increases by 0.01%. Additionally, a 1% increase in the average years of education leads to a 1.80% increase in unemployment, and a 1% increase in inflation causes unemployment to rise by 0.01%.

Regression of gross domestic product with exports, unemployment, and inflation

Table 6 shows the analysis of gross domestic product (GDP) considering structural Equation 6. The results indicate that exports, unemployment, and inflation are significant at the 5% confidence level. The findings explain that exports have had a positive effect on GDP during the analyzed period, as higher exports result in a positive balance of payments, impacting foreign exchange and terms of trade, thereby fostering GDP growth. Conversely, unemployment shows a negative relationship with GDP due to the still high unemployment rates in Peru, which negatively affect economic growth by reducing private investment and capital attraction. Inflation also shows a negative relationship with GDP, as the inflationary spiral leads to a loss of purchasing power, consequently affecting economic growth. This is consistent with Fernando (2016), who evidenced the negative relationship between inflation and economic growth for Honduras, and Uribe (2006), who noted that demand shocks negatively affect GDP and prices in Bolivia, while supply shocks have had a positive long-term impact on GDP and a decline in price levels.

It was detailed that for each 1% increase in the value of exports, GDP increases by 9.5%. Meanwhile, for each 1% increase in the unemployment rate, GDP decreases by 4.3%, and a 1% increase in inflation would cause GDP to decrease by 0.32%.

Effect of public health expenditure on malnutrition by department

In the analysis of the impact of public health expenditure on the reduction of malnutrition by department shown in Figure 1, it is observed that the greatest impact is in the Ucayali department, which reduced the malnutrition rate by 19.17%, followed by Madre de Dios with a reduction of 12.62%, and Ica with a reduction of 12.97%. Conversely, in Pasco, public health expenditure led to an increase in malnutrition by 17.61%, in Arequipa by 15.18%, in Lima by 9.85%, and in Cajamarca by 8.53%. This variability in the impact of public health expenditure on malnutrition during the period 2010-2020 indicates differing efficiencies and effectiveness of public health spending across departments. In 14 departments, including Pasco, Arequipa, Cajamarca, La Libertad, and Lima, public health expenditure did not achieve a reduction in the malnutrition indicator, with the population not meeting the minimum required caloric intake during the period 2010-2020.

Figure 1. Impact of Public Health Expenditure on Malnutrition by Department.

Note. The abbreviations of the regions of Peru are AM (Amazonas), AN (Ancash), AP (Apurímac), AR (Arequipa), AY (Ayacucho), CA (Cajamarca), CU (Cusco), HU (Huánuco), HV (Huancavelica), IC (Ica), JU (Junín), LA (Lambayeque), LL (La Libertad), LI (Lima), LO (Loreto), MD (Madre de Dios), MO (Moquegua), PA (Pasco), PI (Piura), PU (Puno), SM (San Martín), TA (Tacna), TU (Tumbes), and UC (Ucayali).

Figure 2 shows the interrelationship of the variables in the proposed structural equations, where it has been evidenced that public health expenditure has not had the expected impact, as malnutrition levels have increased in Peru. The rural sector shows the most significant relevance with malnutrition due to limitations in accessing adequate food because of low incomes. Thus, the unemployment rate has a positive relationship with malnutrition, demonstrating that the unemployed population would have lower adequate calorie consumption, highlighting a positive association with malnutrition. Meanwhile, the sustained growth of gross domestic product (GDP) produces a negative effect in reducing malnutrition, with the most significant impact being related to the rural population and GDP.

Figure 2. Dynamics of the identified structural equations in the departments of Peru.

Note. Result of the interactions of regressions in the identified structural equations in the departments of Peru

Additionally, it is observed that exports have positively influenced GDP, being considered a pillar of the Peruvian economy’s growth through the diversification of the productive matrix, which leads to a negative relationship with unemployment. Inflation, on the other hand, shows a negative relationship with GDP, as the inflationary spiral causes a loss of purchasing power, thereby affecting economic growth.