1. Introduction

Today's digital world permeates societies and economies around the world to varying degrees. Its economic importance is undeniable. In Mexico, the gross censal added value of e-commerce contributed 5.9% to GDP in 2022 (^{INEGI, 2023a}), while the telecommunications subsector grew at an average annual rate of 9% between 2015 and 2023 (^{INEGI, 2023e}), well above the average annual growth rate, close to 1%. Developed countries perform better globally, with e-commerce value growing at average annual rates of between 8% and 10% in countries such as Germany, South Korea and the United States (^{OECD, 2024}).

In recent decades, specialized literature on the relationship between digitalization variables and economic growth has grown. One World Bank (WB) study (^{Qianget al., 2009}) found that telecommunications variables positively impact on economic growth. ^{Katz's (2009}) analysis validated its positive transversal effect on developed and Latin American (LA) countries through employment, investment, innovation, productivity and economic development. Other studies for Latin American countries have also validated the multiplier effects of technological infrastructure on economic performance (^{Alderete, 2017}; ^{Katz et al., 2023}; ^{Zaballos and López-Rivas, 2012}). For example, ^{Rabanal (2024}) affirms that there is evidence of the technology productivity paradox (^{Solow, 1987}) in his analysis of 12 Latin American countries.

Regarding Mexico, the literature measuring the causal relationship between telecommunications variables and economic growth is scarce. Some studies have been identified, such as that by ^{Jiménezet al.(2014)} with data from 1991 to 2010, which found a positive effect of the variables of mobile phone users and the number of computers on GDP using the Global Innovation Index (GII).¹A subsector study by ^{Díazet al.(2018}) revealed that level of education, information and communication technology (ICT) skills and the availability of ICT infrastructure in companies are important determinants of labor productivity in Mexico. Meanwhile, ^{González (2020}) estimated that the effect of broadband internet increased GDP by 0.13% for the period from 2013 to 2018.

It is important to note the context of this research:i)there is a positive relationship between digitalization variables and economic growth;ii)a greater effect is observed in countries with high levels of economic progress, linked to variables such as human capital, exports, foreign direct investment and public investment (^{Pradhanet al., 2018});iii)the use of individual telecommunications variables or composite digitalization indices has become popular (^{CMD, 2024}; ^{ITU, 2024a}; ^{Miloševićet al., 2018}); andiv)the analyses are intended for countries. This paper goes a step further by constructing a robust digitalization index for the 32 states of Mexico for the period 2015-2023. This index includes socioeconomic variables, ICT assets in homes and ICT skills, in line with international digital development indices (^{EC, 2024}; ^{ITU, 2024b}; ^{Jyoti and Singh, 2023}; ^{Katz, 2009}).

The main objective of the research is to measure the impact of a Digital Economy and Society Index (DESI) on Mexico's economic growth. Dynamic panel regression models are employed, and the impact of digitalization is differentiated according to the states’ economic levels.

The paper is structured as follows: the second section details the conceptual aspects of the relationship between economic growth and technical progress related to digitalization. It then discusses the use of digitalization variables and indices in relation to economic growth. The fourth section details methodological aspects of the DESI and the econometric model specifications. The fifth section presents the econometric modeling and results. This is followed by a discussion and implications. The final section contains the conclusions and annexes.

2. Economic growth and the digital economy: conceptual aspects

^{Schumpeter (1942)} was one of the first to recognize innovation and investment in knowledge and technical progress as sources of economic growth and industrial transformation, capable of maintaining and reproducing the capitalist system. Meanwhile, neoclassical economic theory (^{Solow, 1956}; ^{Swan, 1956}) assumes that economic growth is explained by improvements in productivity and technology rather than directly by the accumulation of physical or human capital. Endogenous growth theorists (^{Aghion and Howitt, 1998} and ²⁰⁰⁸) argue that technical progress can boost economic growth through innovation and knowledge, thus overcoming diminishing returns on capital.

The works of ^{Katz and Koutroumpis (2013) 2}and ^{Pradhanet al.(2014)} posit that technological or telecommunications infrastructure variables embody technical progress and generate increasing returns to scale. Technological progress, digitalization, and the Internet are embodied in countless commercial activities, streamlining government services and connecting suppliers with consumers of goods and services. ^{Tapscott (1996)} defines the digital economy as a combination of networks, knowledge, intelligence and creativity for creating wealth and social development. In the opinion of ^{Kling and Lamb (2000, p. 17)}, “it focuses on goods and services whose development, production, sale or provision depend critically on digital technologies […]". The ^{OECD (2020)}, meanwhile, defines it as involving technologies, digital infrastructure, digital services and data used by three main agents: producers, consumers and the government. Meanwhile, ^{Bukht and Heeks (2017)} speak of a digital economy ecosystem that connects individuals, businesses, devices and society as a whole.

Therefore, the concept of the digital economy encompasses aspects of technological infrastructure, the market and organization. However, it is worth mentioning that it was difficult to measure for many years. ^{Solow (1987, p. 2)} coined the phrase, “The computer age is visible everywhere, except in the productivity statistics," as investments in computers, telephones and technological devices were not reflected in productivity and income statistics.

Therefore, most of the existing literature measures causal relationships between telecommunications variables or digitalization indices to relate them to economic performance, relying largely on official statistics from the World Bank, the World Economic Forum, the International Telecommunication Union (ITU) and the European Commission (EC). In fewer cases, national data is used (^{Chen and Kimura, 2019}; ^{Hanafizadehet al., 2009}; ^{Jyoti and Singh, 2023}). Based on this premise, studies analyzing the relationship between digitalization variables and economic growth were reviewed.

3. Review of the literature on the relationship between digitalization and economic growth

Globally, models have been documented using individual variables of digitalization and economic growth:i)the number of fixed telephone lines or teledensity (^{Datta and Agarwal, 2004}; ^{Norton, 1992});ii)^{Jacobsen (2003)} and ^{Minges (2000)} considered the number of personal computers and the number of telephones, highlighting income, level of education, employment and geographic location as key factors in society and the economy;iii)the WB (^{Qianget al., 2009}) used broadband internet connection, mobile phone use and mobile internet connectivity, respectively;iv)^{Fernández-Portilloet al.(2020}) used internet use and ICT-mediated activity variables (reading digital newspapers, e-commerce and use of e-government).

Digitalization indices are created to measure inequalities or gaps, with a view to transitioning toward the construction of information societies (^{Tello, 2008}). The^{ITU's Digital Development Index (2024b)} shows Mexico above the global average (80.7 versus 74.8). The European Digital Economy and Society Index (^{EC, 2022})³highlights the leadership of the Nordic countries and the United Kingdom.

Similarly, some studies validate the positive effect of digitalization on economic growth in Latin American countries. ^{Alderete (2017)} found that a 1% increase in the number of broadband internet connections would increase average GDP by 0.184%; ^{Zaballos and López-Rivas (2012)} found that a 1% increase in fixed broadband penetration increased GDP by 0.32% in countries in Latin America and the Caribbean (LAC) between 2003 and 2009. ^{Brenes-González (2023}) found that a 1% increase in households connected to the internet increased productivity by 0.04% in Nicaragua from 2000 to 2020. Only ^{Quiroga-Parraet al.(2017)} validated the negative association between low technological penetration and economic backwardness.

The study by ^{Farhadiet al.(2013)}, covering the period 1995 to 2009, used a digitalization index composed of ICTs in developed countries. The results showed that GDP increased by 0.15% for every 1% increase in the index. Using a digitalization index, ^{Katz and Koutroumpis (2013)} found a 0.06% increase in GDP for each percentage point of the index in a sample of 184 countries for the period 2004-2010. ^{Hanafizadehet al.(2009)} used a composite ICT digitalization index for 150 countries, grouped by level of ICT development and validated the leadership of Nordic countries and the lagging of African countries. Meanwhile, the index developed by ^{Gerpott and Ahmadi (2015)} found a greater effect of broadband and mobile telephony variables on aggregate output and development.

Regarding Mexico, cross-sectional studies and studies on the digital divide predominate (^{Escobar and Sámano, 2018}; ^{Martínez-Domínguez and Mora-Rivera, 2020}; ^{Rodríguez, 2019}; ^{Rodríguezet al., 2024}) within a heterogeneous regional context (^{CMD, 2024}). There has been persistent progress in states in the center and north of the country, such as Mexico City, Nuevo León and Baja California, while states in the south and southeast, such as Chiapas, Guerrero and Oaxaca, have fallen behind. This is evident in the DESI calculated for Mexico (see annexes 1, 2 and 3), which shows a structural digital lag in southeastern Mexico.

These studies are linked to this paper through the selected units of analysis (households or ICT users), the sample and the relationship between economic growth and digitalization variables.

In view of the above, the proposal of this paper is based on the following:i)the DESI developed by the ^{EC (2022)}, which focuses on measuring the digital progress of its member countries and guiding policies and strategies for digital transformation and economic growth;ii)the Digital Development Index of the ^{ITU (2024b)}; and the index developed by ^{Farhadiet al.(2013)}, who use this index as a proxy for ICT development in econometric modeling. For Mexico, the DESI proposal reveals the digital evolution of its 32 states and highlights the importance of addressing the level of digitalization (economic and social) at the state level and in public policy on digital matters.

Here, digitalization will be defined as the level of ICT development, estimated by the socioeconomic status of households, the availability of ICT assets and digital skills for specialized internet use. The inclusion of socioeconomic variables is justified by the existence of economic and social inequalities between states, such as poverty, informal employment and low levels of education, among others. The methodological aspects for constructing the DESI are described below.

4. Methodological aspects and data used

In methodological terms, using ICT variables to construct indices is a useful tool when considering a larger amount of information and robustness. The main input for this paper was the processing of microdata from the National Survey on the Availability and Use of Telecommunications in Households (ENDUTIH by its Spanish initials) from 2015 to 2023⁴(^{INEGI, 2015} to ²⁰²³), from questionnaires on dwellings, households and users. With the exception of 2019,⁵the sample is representative at the national and state levels and is comparable for both households and users.⁶Additionally, there is no significant variation between the statistics of some telecommunications variables published by the OECD at the country level and the ENDUTIH statistics.⁷

Unlike the methodologies tested in other studies, such as those of ^{Farhadiet al.(2013)} (eight indicators in three categories: access, use and education and skills indicators), ^{ITU (2024b)} (14 indicators in three categories: access, use and skills), ^{Katz and Koutroumpis (2013)} (16 indicators in six components related to affordability, ICT infrastructure, use and human capital) or the aforementioned DESI for European countries, the selection of socioeconomic and ICT variables (access, use and skills) reflects the use of ENDUTIH microdata and its units of analysis (dwellings, households and users). This allows the index to be constructed over time for the 32 states.

Twenty variables were used to construct the DESI, divided into three categories:1)housing conditions (measuring minimum conditions for basic needs);2)availability of ICT assets and services (approximating minimum conditions for digital inclusion); and3)digital skills (reflecting human capital for interacting in society and the economy). The index was calculated using a simple average, with weightings of 0.20, 0.40, and 0.40, respectively. It is important to take into account the socioeconomic aspects of households, as the index is based on the persistent unaffordability of internet service.⁸

Table 1 lists the variables used to construct the DESI, definitions and weightings. All variables are dichotomous and take the value 1 when the answer is affirmative, at the household or individual level.

The data panel is supplemented by state-level added variables that influence the behavior of the interaction between ICT variables and economic growth behavior, namely public investment, exports and foreign direct investment (^{Datta and Agarwal, 2004}; ^{Katz and Koutroumpis, 2013}; ^{Pradhanet al., 2018}) (see Figure 1). The result was the creation of a data panel with 288 observations (32 states and nine years) and a programming code in the Stata version 17 package.

Notes: DESIMEX (Digital Economy and Society Index of Mexico); the dependent variable is real GDP in millions of pesos (for convenience, GDP). Boxes 1, 2 and 3 list the variables used to construct the DESI: housing conditions, ICT availability and digital skills. The economic variables are listed in the lower right-hand corner. GFCF: gross fixed capital formation (millions of pesos) or public investment; EXP: exports (millions of USD); FDI: Foreign Direct Investment (millions of USD); individual ICT variables: internet user inside or outside the home in the last three months (IU), mobile phone user (MU), internet service at home (IS) and computer at home (DC) and the DESI (DESIMEX).

Source: prepared by the author using the ^{Mermaid tool (2025)} and information from ^{ENDUTIH-INEGI (2015} to ²⁰²²; ^2023b), ^{INEGI (2023c}, ^2023d, ^2023e) and the ^{Ministry of the Economy (2023)}.

Figure 1 ICT and regional variables influencing economic state growth, 2015-2023 

States were classified as having high or low GDP, based on real GDP per capita in 2023 (see Annex 4),⁹ was made to measure the differentiated effect of digitalization variables on economic growth (^{Abramovitz, 1986}; ^{Diazet al., 2016}; ^{Singh and Kumari, 2023}) or what ^{Koutroumpis (2009)} refers to as network externalities, in the sense of a non-linear effect of ICTs on growth, resulting in higher than proportional returns when certain levels of critical mass are reached. The values of aggregate variables in each groups of states (see Annex 5) reflect the different stages of economic and digital development. For instance, the variable of internet use is presented as an initial phase, while variables such as computer availability in the home or the DESI variable indicate a higher level, related to specialized internet use.¹⁰

5. Econometric modeling and results

Due to the non-contemporary relationship between digitalization variables and economic growth, dynamic panel models are proposed. As a result of the structure of the data and the context of the analysis (states within a country rather than countries), two regression techniques could be tested: first, the GMM difference method (^{Arellano and Bond, 1991}) and, second, the GMM system method (^{Blundell and Bond, 1998}; ^{Roodman, 2009}). The first model eliminates unobserved fixed effects and uses lags of the endogenous and exogenous variables as instruments to eliminate endogeneity issues (^{Ruiz-Porras, 2012}). Both types of models adjust aspects of panel data, serial correlation, heteroscedasticity and endogeneity.

The GMM system model allows us to overcome the shortcomings of the traditional ^{Arellano and Bond (1991)} model. One such case is when the instruments are weak or invalid. In this case, the GMM system model can combine variables in levels and differences to solve the problem of potentially endogenous explanatory variables and overcome heteroscedasticity problems through more robust estimates.

For convenience, all variables were transformed into logarithms to facilitate the analysis and interpretation of results in terms of elasticities and isolate the instantaneous effect of each independent variable on the dependent variable. The canonical dynamic panel regression model in levels is as follows:

Where:

Due to the reduced number of periods and the relatively small number of groups, the model type with exogenous independent variables was not used, i.e., lag operators were not used for these variables in order to have greater control over the number of instruments (^{Mileva, 2007}). Previously, the procedure was carried out to validate the panel data model as the best model specification, i.e., the Breusch-Pagan Lagrange Multiplier (BPLM) test was run to determine the presence of a panel effect. In all three cases, the null hypothesis was rejected (see Table 2).

Six regressions are run using dynamic panel specifications. The first three use dynamic models of the differenced GMM estimators type are run, which are particularly useful when there are many groups and few periods and when exogenous, predetermined, and/or endogenous independent variables can be included (^{Labra and Torrecillas Bautista, 2018}). The remaining three use dynamic GMM system models were run in order to reaffirm the results and use more variables as instruments (^{Roodman, 2009}).

The technique developed by ^{Arellano and Bond (1991}) addresses endogeneity and multicollinearity problems through the use of instrumental variables.

The functional form of the dynamic model of the GMM system to eliminate individual effects (h i) is as follows:

To obtain differentiated results in the coefficients for each sample (all states, states with high GDP and states with low GDP) and based on references in the literature (^{Farhadiet al., 2013}; ^{Gerpott and Ahmadi, 2015}; ^{Jyoti and Singh, 2023}), we analyze the DESI index variable and the four individual TIC variables simultaneously.

Substituting the variables, the regression is as follows:

Where:

GDP_i,t = Dependent variable. Real GDP at 2013 prices.

αGDP_{i, t-1} = Term representing the memory effect or dependence on the past.

DESI = Digital Economy and Society Index, composed of 20 variables in three categories.

IU= Internet user, at least once in the last three months (user questionnaire).

MU= Mobile phone user, at least once in the last three months (user questionnaire).

SI= Internet connection at home (household questionnaire).

DC= Home computer, desktop or laptop (household questionnaire).

IP= Public investment or gross fixed capital formation. In Mexican pesos.

EXP= Value of merchandise exports, resulting from Mexico's trade with the rest of the world. In USD.

FDI= Foreign Direct Investment flows at the end of each year. In USD.

d_2020 = Allows us to capture the effect of the fall in GDP due to the Covid-19 pandemic.

α₀= Constant. A value is obtained for each of the three samples used.

ε_i,t = Error term representing the unexplained variability in the model.

A positive effect is expected for all explanatory variables. Subindexiconsiders the 32 states in the first two regressions; 16 states with high GDP in the next two regressions; and 16 states with low GDP in the last two regressions.

The GMM system model allows moment conditions to be added at different levels. In general, this model is ideal because:1)it allows two estimators to be calculated, the GMM difference estimator and the augmented estimator or GMM system (^{Bond, 2002});2)more efficient estimators are achieved in two stages than in one stage and it allows for finite sample correction of the two-step covariance matrix (^{Windmeijer, 2005}), achieving reliable and robust estimators; and3)it eliminates the problem of instrument proliferation.

The expressions of the two equations of the GMM system panel data model are:

Similarly, when substituting the variables in the model, the equation expression reads as follows:

After substituting the variables and specifying the GMM system panel data model (^{Roodman, 2009}),¹¹ the equation was run in Stata software. The results of both specifications are presented in Table 3.

6. Discussion and implications

Dynamic panel econometric modeling is ideal for overcoming problems of endogeneity, multicollinearity and serial correlation in samples with many cross sections and few periods. The effect that the DESI would have on economic growth in Mexico is tested.

The main finding of the model is the explanatory power of the DESI on economic growth. Considering all states, a 1% increase in this index boosts national GDP by 0.054%-0.058% (ceteris paribus); a similar effect to the 0.064% found by ^{Rabanal (2024)}, who used the percentage of households with an internet connection and the 0.06% found by ^{Katz and Koutroumpis (2013}), who used a digitalization index. For the group of states with high GDP, the effect of the DESI is more than double, at 0.134%. This value is slightly higher than the WB estimate for developed countries (^{Qianget al., 2009}) of 0.121%, which uses broadband internet connectivity as a variable, slightly lower than ^{Alderete’s (2017)} estimate of 0.184% and similar to the value found by ^{Farhadiet al.(2013)} of 0.15%. The results also align with ^{Singh and Kumari’s (2023)} findings of only positive effects of telecommunications variables in the group of rich countries, with an increase in GDP of 0.08% and 0.06%, using variables such as landline telephone subscriptions and internet users, respectively.

In states with low GDP, the DESI has no effect on economic growth. This is consistent with the argument that a critical mass is needed to take advantage of digitalization (^{Koutroumpis, 2009}; ^{Mariscal, 2022}). The mobile phone user variable confirms that Mexico has surpassed a certain stage of development as there is a negative effect on economic growth, especially among rich states, although the provision of quality services and non-concentrated market structures remain pending (^{Mariscal, 2022}).

Interestingly, the effect of the computer availability variable is positive for the dynamic system model specifications and that the effect is greater among poor states. A 1% increase in the DESI increases GDP by 0.233%. This latter result contrasts with that of ^{Jacobsen (2003}), who found a positive effect of personal computers in developed countries from 1990 to 1999, but not in developing countries. Jacobsen considered personal computers to be necessary goods in developed countries and luxury goods in developing countries. The idea that computers are non-essential goods is consistent with the percentage of Mexican households with computers, which fell from 44.9% to 43.8% between 2015 and 2023 (INEGI 2015 and 2023). This result appears to contradict the effect of the COVID-19 pandemic on the intensive use of the internet, technology and electronic devices.¹²

Another noteworthy item is the effect of aggregate variables on economic growth. An increase in public investment invariably has a positive effect on economic performance, especially for states with a high GDP compared to those with a low GDP. Regarding exports, this effect only applies to rich states, where a 1% increase in public investment raises GDP by 0.063% to 0.066%, consistent with the industrial, manufacturing and export profile of these states. The null effect in states with a low GDP indicates the pattern of economic growth and historical growth strategy, focused on foreign trade and investment, as opposed to the lagging states in the south-southeast.

Regarding the validity of the models, first, Wald's statistics indicate that the coefficients of the explanatory variables in the model are statistically significant overall, i.e., they impact GDP behavior (a weak argument since it indicates that at least one of the coefficients has a value different from zero). Secondly, the number of instruments must be less than the number of groups. Only models with the entire sample meet this condition and the data panel used negatively influences it. Dividing the sample into two automatically reduces the number of groups. Furthermore, it is not appropriate to validate the Sargan test in this type of model since the use of instruments causes over-identification of the model.¹³ Thirdly, the goal is to ensure that there is no second-order serial correlation in the residuals (AR 2). The null hypothesis that there is no autocorrelation was not rejected in all specifications so the models are valid.

The results of the econometric modeling reveal the possible path to be followed by public policymakers in terms of telecommunications investments and sector promotion. For low-income states, the greatest multiplier effect would come from increasing the availability of computers (in excess of those documented in literature), public investment and exports. For high-income states, digitalization would generate greater economic dividends in terms of aggregate income, employment and industrial development- For economically lagging states, the most pressing strategy would be to invest in physical infrastructure and connectivity (means of transport) and encourage productive vocations in the service sector.

7. Conclusions

A data panel was constructed for Mexico, broken down by state from 2015 to 2023. This allowed us to investigate the causal relationship between an economic and social digitalization index and economic growth. A positive and statistically significant effect was validated: a 1% increase in the DESI boosts total GDP by 0.058%. This effect doubles when considering the sample of states with high GDP, reaching 0.134%. Consistent with the reviewed literature, relating economic growth to digitalization using information from households or individuals as the unit of analysis is feasible. Likewise, the findings confirm and suggest the following:i)the leadership of states with greater economic progress and the potential use of digitalization as an industrial policy instrument;ii)the close relationship between economic performance, digitalization and human capital; andiii)the validation of stylized facts on the positive and differentiated impact of digitalization variables on economic growth; andiv)the importance of focusing on the issue of digital skills to maximize the potential of connectivity, provided there are institutional capacities, physical and technological infrastructure and market regulation.

Based on a review of the literature and studies conducted for Mexico, this research is novel in the following ways:i)it measures and analyzes a digital economy and society index over time for the 32 states. This is commonly carried out for groups of countries by international organizations (^{ITU, 2024b}; ^{EC, 2022}). In Mexico most studies using ICT-related indices are cross-sectional (^{Escobar and Sámano, 2018}; ^{Martínez-Domínguez and Mora-Rivera, 2020}; ^{Rodríguez, 2019});ii)it empirically validates the impact of a digitalization index on economic growth. It does so using microeconomic and secondary source information. It differentiates the effect according to the economic level of the states, unlike studies that use individual ICT variables (^{Figueroa Hernándezet al., 2021}; ^{González, 2020}; ^{Rabanal, 2024}); andiii)positioning the close relationship between ICTs and economic growth as an area with real potential to boost the national economy.

It should be noted that this study had two limitations: the lack of information on household income or expenditure, which was unavailable in the ENDUTIH and the absence of human capital variables such as tertiary education, investment in science and technology and productivity. It is important to note that the analysis was carried out during a period of economic slowdown, when the average annual growth rate of real GDP was 0.28% and real GDP per capita was -1%. These elements should be considered when designing an ambitious and long-term digital industrial policy. There is still room for further analysis of the digitalization process in Mexico at the regional or territorial level, taking into account productive, scientific and resource allocation vocations. This research therefore opens the door for future research on the impact of the digital economy in the economic and social spheres.

Clearly, the reduction in spending on fixed internet connectivity and computers in Mexican households is due to the economic situation and the economic slowdown. Therefore, any strategy to promote connectivity in the digital age must involve designing subsidy programs for connectivity, increasing the development of freely accessible content and promoting collaboration between private entities and the government. A new public policy on digital matters must be developed that focuses on the digital economy and responds to the challenges of a hyper-connected society.