Dynamics of Youth and Gender Divide in Technology in Bangladesh

Abstract

This article identifies the nature of gendered digital divides between male and female youth (aged 15–29) in the context of Bangladesh. As a measuring indicator for technology inclusiveness, this study examines whether a young male or female owns a mobile phone, the most basic means of accessing the Internet. As observed in the descriptive analysis, on average, 46% of young females have a mobile phone, compared to 79% young males. However, such disparity varies across age cohorts and the divides of rural–urban, poor–non-poor, richer–poorer income deciles, etc. To understand whether there is any significant discrimination against women in terms of technology inclusiveness, this study applies the Blinder–Oaxaca (B–O) decomposition technique. The decomposition analysis shows statistically significant discrimination against women in terms of mobile ownership at both the household and the individual level. The factors such as remittances, average years of schooling of the household members, urban residence, household’s income status, etc., significantly reduce discriminatory behaviour towards young females.

JEL: D63, O33, I21

Keywords

Digital divide Blinder Oaxaca decomposition technological inclusiveness youth gender discrimination

Introduction

Technological innovation is considered one of the fundamental ‘engines of progress’ of world development (Crewe & Harrison, 1998). The transfer of technologies from the Global North to the Global South has been considered as one of the key channels for social and economic development. There is no denying that technological innovation has uplifted many of the socio-economic indicators of emerging countries. In other words, in the context of the 21st century, progress in technologies can be considered as a necessary condition for sustainable development.

However, the experience of technological innovation is not age- and gender-neutral. Compared to the developed countries, emerging countries like Bangladesh have fewer opportunities and resources to prepare the youth with adequate training for facing the challenges of the current and future job market. Youth left out of technological innovations will be unable to meet the challenges of the Fourth Industrial Revolution (FIR). Since youth comprise a significant portion of the population of Bangladesh, such unreadiness might result in adverse impacts on sustainable development.

Nonetheless, the gender divide in technology is more common. In many developing countries, such as Bangladesh, technology’s dividends are not equally distributed between males and females. Compared to men, women have lower access to technology due to a lack of skills or digital literacy, social stigma or other socio-economic barriers. Such access to technology ultimately deteriorates the pre-existing power imbalances in the country (Sweetman, 1998). If the gender perspective of technology inclusiveness is not adequately addressed, the Sustainable Development Goal (SDG) of ‘Leaving No One Behind’ will remain unattained. Without appropriate measures, the progression of the country towards FIR could expand the gender gaps in various socio-economic frontiers. Therefore, when aiming to attain gender parity in development and thus SDG 5, equality in access to technology is pertinent.

Noting the importance of technological readiness, the Government of Bangladesh (GoB) rightly adopted the ‘Digital Bangladesh’1 agenda in 2009, marking it as a national priority (Islam & Tsuji, 2011). Although the GoB had a solid intention in marking the initiative a success, due to several structural loopholes, the agenda was not as successful as promised. There are concerns around whether access to technology has been equally ensured across genders, age cohorts, regions (rural/urban) and income groups. No study in the context of Bangladesh has explored the nature and extent of technological inclusivity in Bangladesh.

Against this backdrop, an analytical investigation on the current scenario of youth and gender inclusivity and accessibility to technology in Bangladesh could not be timelier. This article identifies the nature of the gendered digital divide between young males and females (aged 15–29) in Bangladesh. As a measuring indicator for technology inclusiveness, this study examines whether a young male or female owns a mobile phone, the most basic means of accessing the Internet. The study conducts a detailed descriptive analysis to understand the pattern and nature of the relationship between the mobile holding status of the youth by gender and the factors such as rural–urban divide, poverty, differences in income deciles, difference in age cohorts, etc. To understand if there is any significant discrimination against women in this regard, this study applies Blinder–Oaxaca (B–O) decomposition and an extension of B–O decomposition coined by Fairlie (1999).

The article is designed as follows. After this introduction, the second section details the background of this study. The third section outlines the methodology adopted in this study. The fourth section provides some relevant descriptive statistics on the subject matter of the study. In the fifth section, the study investigates gender discrimination in technology among youth, empirically applying B–O decomposition. Finally, the sixth section provides concluding remarks.

Background and Literature Review

As a starting point, it is essential to explore why technology inclusiveness matters for development, define what is meant by technological inclusiveness (or digital divides) and explain how to measure it.

The world is now approaching FIR. It is driven by an information economy where computers and the Internet play a significant role (Castelles, 2004). FIR is characterized by advances in artificial intelligence (AI), robotics, the Internet of Things, three-dimensional (3D) printing, genetic engineering, quantum computing and other technologies (Schwab, 2016). With each industrial revolution, the world passed through a significant structural transformation with long-lasting impacts on production techniques and consumer preferences (Warschauer, 2003).

FIR will have a long-term impact on efficiency and productivity on the supply side. Like the three previous industrial revolutions, FIR is also expected to raise global economic growth or income levels and improve the quality of lives. However, with advances in technologies and AI, there are also threats of massive job cuts and rising inequalities. Poor and less skilled workers (early youth, the elderly and women) will face greater threats than others. With the gap in earnings between skilled and unskilled workers being widened, social inequality might also see a rise. With females being less skilled than males, such a change in earnings will widen the gender-based wage gap between males and females. Given this context, the only way to reap the benefits of FIR is ensuring technology inclusiveness of the future and current workforce (mostly the youth) of the country.

In this era of information technology, technological inclusivity is closely related to access to information and communications technology (ICT). The simplest, but perhaps the most conservative, way to think about ICT access could be the ownership of an ICT device (such as a computer, internet connection, mobile phone, etc.). In this sense, technological inclusiveness can be defined as access to such ICT devices. It must be noted that mere access to such ICT devices does not portray the actual scenario of ‘technological inclusiveness’. Proper technological inclusiveness would require connection to the Internet and the skills and understanding to use the Internet in socially valued ways (Warschauer, 2003).

Digital Divide and Its Background

The societal split in the case of technology is commonly referred to as the digital divide. The term digital divide was first coined by the US National Telecommunications and Information Administration in the mid-1990s. The digital divide refers to the separation between those who have access to digital ICT, such as the Internet or computers, and those who do not. In the literature, the concept of a digital divide has been used to comprehend technology’s relationship with equity and development. The digital divide can emerge from the existing disparities in terms of access to technology across different income groups, races or genders.

The global literature on the digital divide can be divided into two categories. While much of the study on the digital divide focuses on the first-order effects regarding who has access to the technology (mobiles, the Internet or computers; Chakraborty & Bosman, 2005; Prieger & Hu, 2008; Rice & Katz, 2002; Tran et al., 2015), some works address the second-order effects of inequality in the ability to use technology among those who do have access (Cooper, 2006; Jones et al., 2009; Keegan, 2004; Livingstone & Helsper, 2007; Venkatesh et al., 2000).

Keegan (2004) examined the digital divide among the poor and non-poor youth of the USA in terms of the divide in home computer ownership and the differences in academic and non-academic computer use. They found that youth from low-income families were likely to use their home computer for academic purposes but less likely to use any computer for non-academic purposes.

Cooper (2006) examined the evidence for the digital divide based on gender. In this study, the gendered digital divide is measured based on the differences between males and females in learning to use a computer and learning other material with the aid of computer-assisted software. The study showed that females are at a disadvantage relative to men in learning to use a computer and learning other material with the computer.

It is noteworthy that the digital divide concerns not only material access but also skills and usage patterns (Fuchs & Horak, 2008). Although material access is necessary, it is not a sufficient precondition for skills access and user access. Fuchs and Horak (2008) pointed out the information society indicators used by the European Union, some of which also focus on skills and usage. One of the indicators measures the percentage of individuals using the Internet for specific purposes (sending/receiving emails, finding information about goods and services, reading/downloading newspapers, internet banking) in the previous 3 months.

Korupp and Szydlik (2005) analysed the concept of digital divide by studying the causes of private computer and internet access with a threefold model including human capital, family context and social context. They found human and social capital more important than economic capital in explaining private computer and internet use. Rice and Katz (2002) analysed three kinds of digital divides between internet and mobile phone users and non-users, users and dropouts and recent and veteran users. Their findings show that the three kinds of digital divides are conceptually and empirically different, both with and across internet and mobile phone media.

Livingstone and Helsper (2007), using a national survey of the United Kingdom comprising participants aged 9–19, investigated the digital divide among children and young people. They considered internet access as an indicator of technological inclusivity. Instead of looking for a binary divide, they analysed the digital divide with continuous indicators, such as the frequency of internet use (from low users and non-users to weekly and daily users). The study reveals inequality by age, gender and socio-economic status in relation to their quality of access to and use of the Internet.

In the context of rural Bangladesh, Tran et al. (2015) examined the demographic determinants of household mobile ownership using socio-economic data collected as part of a multi-year longitudinal cohort study of married women of reproductive age. They explored how the demographics of household mobile phone owners have changed over time in a representative population of rural Bangladesh. The data were collected between 2008 and 2011 on household mobile phone ownership and related characteristics, such as age, education, electricity access, employment and household wealth. They found that household ownership of at least one mobile phone grew from 29.85% in the first fiscal year to 56.06% in the third fiscal year. The study identified illiteracy, unavailability of electricity and low quartiles of household wealth as overall constraints to mobile phone ownership.

Zhou et al. (2011), using survey data from three South Asian countries (Bangladesh, Nepal and Sri Lanka), investigated how low-cost Internet can be provided with the existing conditions of income levels through organizational innovations that help to bridge the digital divide. They examined the factors influencing computer and internet use patterns in these countries and found education playing the key role in terms of its acquisition as a reason for computer and internet use.

The present study considers the gender divide in terms of mobile phone ownership among the male and female youths of Bangladesh. No existing study has explored the gender divide following standard decomposition techniques.

Given that the most comprehensive and detailed (at the individual level) data are available only for mobile ownership (from the HIES (2016)), this study incorporates that indicator for detailed analysis from income, regional and gender perspectives. It is assumed that minimum access to modern technology services is closely linked to ownership of a mobile phone, particularly for females. This is the minimum that a female would require to use the Internet, or join training sessions, or reap the benefits of ICT. Against this backdrop, mobile ownership of the youth has been used to analyse the digital divide based on gender.

Methodology of the Study

Data Source

The study has used the Household Income Expenditure Survey (HIES) 2016. HIES is a cross-sectional survey conducted by the Bangladesh Bureau of Statistics (BBS). This is the largest nationally representative household-level data set of Bangladesh, which contains detailed information on household and individual characteristics, such as income, expenditure, consumption, savings, education, health and sanitation, electricity, household assets, migration, etc. HIES (2016) contains 186,000 observations from 46,076 households.

Empirical Methodology

The main objective of decomposition analysis is to observe the inter-group differences in the mean level of outcomes into those originating from observable characteristics (such as education) and those arising from other characteristics (such as gender or race). B–O decomposition is a widely used decomposition technique (Blinder, 1973; Oaxaca, 1973). It categorizes the outcome variable into two parts: a part that can be explained by the differences in observed characteristics and a part that is attributable to the coefficient (due to discriminatory behaviour).

With this technique, it is possible to decompose the differential in outcomes (such as mobile holding) between two groups into a part that is explained by the group differences in characteristics and a residual part that cannot be explained by the differences in such determinants (Ben, 2008). This unexplained part is often considered as a measure of discrimination. However, it also contains the effect of group differences in unobserved characteristics. Given this context, the study employs the B–O decomposition technique to analyse if there is any gender discrimination in the case of the technological inclusivity of youth.

Due to the unavailability of data on computer or internet access by gender, in this study the status of mobile ownership has been used as an indicator of technology inclusiveness.

This study uses two outcome variables: (a) household per capita mobile ownership of youth calculated as the ratio of the number of youth owning a mobile to the total number of youth in the household (by gender); and (b) youth mobile ownership—a dummy variable that takes the value of 1 if the youth (aged between 15 and 29) owns a mobile and 0 if the youth does not own a mobile. This study compares gender, indicating whether the youth is a male or female.

The following linear regression model fitted separately for males and females (Ben, 2008).

Y_{i}_{m} = β_{m} X_{i} + μ_{i}_{m}

(1)

Y_{i}_{f} = β_{f} X_{i} + μ_{i_{f}}

(2)

where m denotes male youth, f denotes female youth, Y is the outcome variable and X is a vector of predictors indicating household and individual characteristics, such as the education of the youth, age of the youth, etc.

After taking expectations of both sides of Equations (1) and (2) and carrying out some manipulations, we obtain the mean outcome differences as the differences in linear prediction evaluated at the group-specific mean of the regressors as follows:

R = E (Y_{m}) - E (Y_{f}) = E (X_{m}) β_{m} - E (X_{f}) β_{f}

(3)

Equation (5) can be rewritten following Oaxaca and Ransom (1994) for a threefold decomposition as follows:

\bar{Y_{m}} - {\bar{Y}}_{f} = (\bar{X_{m}} - \bar{X_{f}}) β_{f} + {\bar{X}}_{f} (β_{m} - β_{f}) + (\bar{X_{m}} - \bar{X_{f}}) (β_{m} - β_{f})

(4)

Δ \bar{Y} = E + C + I

(5)

Here, the group differences in predictors (Y) are written with respect to females.

The first term of this decomposition shows the part of the difference related to group difference in observable covariates or endowments (E) and is weighted by a vector of coefficients of females. The endowment component (E) measures the expected change in females’ mean outcome if they had the value of explanatory variables of males or if they had males’ predictor levels. The term ‘C’ shows the part of the divide arising from the group differences in coefficients. This coefficient term (C) measures the expected change in females’ mean outcome if they had the same coefficient as that of males.

Finally, the term I denotes the fraction of the gap between the interaction of the group differences in endowments and coefficients between the two groups. In other words, the interaction term shows the gap that occurs when both endowments and coefficients change simultaneously. The sum of the coefficient and interaction term shows discriminatory behaviour. The higher the endowments, the lower is the discriminatory behaviour, and vice versa.

For household-level decomposition, the outcome variable is the household per capita mobile ownership of youth, which is continuous. Therefore, the digital gender divide can be decomposed using the simple ordinary least squares (OLS) under the B–O decomposition technique.

However, when the digital gender divide of youth is decomposed at the individual level, the outcome variable is binary, indicating the mobile ownership of the youth. When the outcome variable is binary, the simple OLS will provide a biased estimate. In this case, the coefficient estimate should come from a logit or probit model. However, such non-linear models cannot be directly used in the conventional B–O decomposition technique (Sinning et al., 2008). Although Sinning et al. (2008) presented the detailed technique of performing the B–O decomposition in such case, Fairlie (1999) first described a relatively simpler way of performing the decomposition applying non-linear models.2

This study follows the approach proposed by Fairlie (1999) to obtain the decomposition results at the individual level to identify how many of the gender differences are due to the group differences in explanatory variables and how many are caused by the discriminatory behaviour. The decomposition technique proposed by Fairlie (1999) for non-linear regression is a twofold decomposition, unlike the threefold decomposition described earlier for the household level. One advantage of using the technique proposed by Fairlie (1999) is that this decomposition technique provides an estimate of the contribution of gender differences in the entire set of the explanatory variables to the gender gap in mobile ownership. To obtain the differences in the expected probability of owning a mobile, two separate regressions are estimated by probit models.

As noted earlier, Equation (3) shows the B–O decomposition of the male–female gap in terms of the average value of household per capita mobile ownership of youth (the outcome variable) using the linear regression technique. However, for the individual-level decomposition, Equations (1) and (2) are non-linear. A non-linear equation such as Y = F (X β) the decomposition equation following Fairlie (1999) can be written as follows:

\bar{Y_{m}} - \bar{Y_{f}} = [\sum_{i = 1}^{N^{m}} \frac{F (X_{I}^{m} β^{f})}{N^{m}} - \sum_{i = 1}^{N^{f}} \frac{F (X_{i}^{f} β^{f})}{N^{f}}] + [\sum_{i = 1}^{N^{m}} \frac{F (X_{i}^{m} β^{m})}{N^{m}} - \sum_{i = 1}^{N^{m}} \frac{F (X_{i}^{m} β^{f})}{N^{m}}]

(6)

where N^m is the sample size of male youth and N^f is the sample size for female youth. This alternative method of representing the decomposition equation has been used because $\bar{Y}$ does not necessarily equal $F (\bar{X} β)$ .

The first term in the brackets of Equation (8) represents the part of the gender gap in case of technological inclusivity of youth which is due to the group differences in the distribution of the explanatory variables, and the second term represents the part of the gap that is due to differences in group processes determining the level of Y. This second term also represents the gender gap due to group differences in unobserved endowments or the gender gap arising from discriminatory behaviour against females. The first term also gives an estimate of the contribution of the gender differences in the entire set of independent variables to the gender gap in the mobile ownership of youth. However, it is also possible to identify the contribution of group differences in a specific variable to the gender gap. If we assume that X includes two variables X₁ and X₂, the independent contribution of X₁ to the gender gap is expressed as:

\sum_{i = 1}^{N^{f}} F (X_{1 i}^{m} β_{1}^{f} + X_{2 i}^{f} β_{2}^{f}) - F (X_{1 i}^{f} β_{1}^{f} + X_{2 i}^{f} β_{2}^{f})

(7)

This individual contribution of each variable to the gender gap is calculated only for the endowment effect, but the contribution to the unexplained part or discriminatory part is ignored due to the difficulty in explanation (Fairlie, 1999).

Descriptive Statistics

For taking appropriate policy measures to ensure technological inclusivity of youth, there is no alternative to taking stock of the existing situation based on data. Such data analyses provide an in-depth understanding of the current gaps in outcomes, help assess the needs of specific pockets (such as lagging regions), and identify specific policy options.

Brief Profiling of the Youth by Broad Age Categories

A ‘youth’ has been defined in this study as anyone in the age group of 15–29. This is the same definition used in the Labour Force Survey of Bangladesh conducted by BBS. Nevertheless, such a broad classification of youth might not provide adequate representation of youth’s inclusivity in technology. To further broaden the scope of the analysis, the youth in this study are grouped into three broad categories: (a) early youth (aged between 15 and 19); (b) middle youth (aged between 20 and 24); and (c) late youth (aged between 25 and 29).

Overall, youth (15–29) comprise more than a quarter of the total population, of which 53% are females (22 million) and 47% are males (19.7 million). According to the broad classification, 37% of the youth fall in the early-youth category (15–19), 30% fall in the category of middle youth (20–24), and 33% fall in the category of late youth (25–29) (Table 1).

Table 1.

Profile of the Youth by Age Categories.

Youth Categories	Gender	Population (in millions)	% of the Total Population	% of the Total Youth
Early youth (15–19)	Male	8.1	5	19
	Female	7.3	5	18
	Total	15.5	10	37
Middle youth (20–24)	Male	5.5	3	13
	Female	6.9	4	17
	Total	12.4	8	30
Late youth (25–29)	Male	6.1	4	15
	Female	7.8	5	19
	Total	13.9	9	33
Overall youth (15–29)	Male	19.7	12	47
	Female	22.0	14	53
	Total	41.7	26	100

Source: Authors’ calculation based on HIES (2016).

Household Per Capita Mobile Ownership of Youth and Adult by Sex and Location

One approach to observing the nature and extent of the digital divide by region and by gender could be the average number of mobile owners among youth by gender in a household. For instance, consider a hypothetical household with three male youth and three female youth members. If all the young males have a mobile phone in the household, that would mean the per capita mobile ownership of the youth male in the household is 1 (or 100%). In the case of females, if only one female holds a mobile phone in that household, then it would indicate that the average number of mobile owners among the young females in that household is 0.33 (or 33% of the females has a mobile phone). Any difference in per capita mobile ownership in households by gender and location might indicate systematic gender bias.

Based on HIES (2016), it is observed that the average mobile ownership per capita for male youth in a household is significantly higher (79%) compared to females (46%) (Figure 1). The same thing is observed for adult (Figure 2). The status of mobile ownership for males does not vary across regions. For instance, on average, in a rural household, 78% of the young males have a mobile phone, compared to 82% young males in an urban household. However, in the case of females, only 42% of the youths have a mobile phone in a rural household compared to 57% in an urban household.

Household Per Capita Mobile Ownership by Income Decile

The household per capita mobile ownership of youths and adults varies across locations. To some extent, these regional differences can be explained with the aid of differences in educational outcomes, the concentration of manufacturing activities (such as ready made garment (RMFs)), social norms and other socio-economic factors. However, it is vital to see whether the pattern varies with income differences as observed in the literature (Chinn & Fairlie, 2004).

When the average per capita mobile ownership at the household level is observed based on income deciles, a clear relationship is observed between household income and gender equity (Figure 3). Households at the upper-income deciles (measured with the per capita total expenditure of the households) have more gender parity in terms of technology inclusiveness than the households at the lower-income deciles. The male–female gap among youth in technology accessibility is 44 percentage points in the first income decile, while the gap is 19 percentage points in the richest income cohort. A similar pattern is observed in the case of adult mobile ownership (Figure 4).

The Intra-household Gendered Digital Divide Among the Youth

The above discussion portrays the divide in technology inclusivity in Bangladesh by income, gender and location. However, the discussion does not include intra-household inequality in technology inclusivity. Therefore, this study incorporates an intra-household gender parity in the technology index.

The score of the intra-household gender parity index is calculated as the household per capita mobile ownership of female youths divided by the household per capita mobile ownership of male youths,3 that is, intra-household gender parity index.

Intra-household gender parity index

= \frac{\frac{N u m b e r o f f e m a l e y o u t h w i t h m o b i l e i n a h o u s e h o l d s}{T o t a l n u m b e r o f f e m a l e y o u t h i n t h a t h o u s e h o l d}}{\frac{N u m b e r o f m a l e y o u t h w i t h m o b i l e i n a h o u s e h o l d}{T o t a l n u m b e r o f m a l e y o u t h i n t h a t h o u s e h o l d}}

If in a typical household there are three female youths and three male youths where only one female youth has a mobile phone and all male youths have a mobile phone, then the per capita mobile ownership among the female youths in that household will be 1/3 (0.33 or 33%), and for the males it will be 1 (or 100%). The ratio of these two values (i.e., 0.33, in this case) would indicate the nature of intra-household gender parity for access to ICT technology. A score of 1 would indicate perfect parity for a household, a score higher than 1 would indicate a female-biased household, and a score lower than 1 would suggest a male-biased household. The closer the value of this score is to 1, the more gender-equal the household is in terms of per capita mobile ownership of the youth. A score of 0.50 for a household can be interpreted as, on average, half the female youths in the household having a mobile phone compared to the male youths.

The intra-household gender parity in mobile ownership varies with the location of the households, as shown in Figure 5. The scores indicate that intra-household gender parity in technology is higher in urban households compared to rural households.

The intra-household gendered digital gap might also depend on the income status of the households. To discern the pattern, the households are divided into 10 quartiles based on household income. As observed, the gender gap in technological inclusivity is the highest for the households from the lowest income decile with a score of 0.28 (Figure 6). An upward trend of the score implies that the intra-household gender parity score improves as the total per capita income of the household increases. The gender parity is highest for the top income decile with a score of 0.74. A similar pattern is observed for the poor and non-poor households. The average intra-household parity score for the poor households is 0.30, while for the non-poor households it is 0.48 (Figure 7). This relationship indicates that income and poverty are important indicators of intra-household gender parity in technological inclusivity.

Technological Inclusiveness by Different Youth Age Categories

The anecdotal analysis provides sufficient justification that there is a digital divide among youth by gender, income and location at the household level. However, it still lacks a complete picture of the technological readiness of the youth by different age categories within the overall youth. In this respect, the measures of per capita mobile ownership can be dissected further by the broad youth age categories, namely early youth (15–19), middle youth (20–24) and late youth (25–29).

With a rise in age (from early to late youth), the percentage of youth owning a mobile phone increases (Figure 8). However, the pace at which it increases is significantly different between males and females. For instance, in the case of early youth, on average, 62% of the males have mobile phones, whereas among the females this rate is only 31% (Figures 8 and 9). In the case of middle and late youth, more than 90% of the males have a mobile phone, whereas in the case of females, the rate is around 53%. In the case of males, there is no significant difference in the rate of mobile holding for the youth across the divisions or rural/urban by the broad age groups. However, in the case of females (Figure 9), a clear regional disparity is observed for all age categories. Across all the age categories, urban youth have a higher average mobile ownership than rural youth, and Dhaka and Chattogram have higher rates compared to other regions. For all age categories and all regions, mobile ownership among the females is lower than that among the males.

Technological Inclusiveness by Different Youth Age Categories (By Gender and Poverty Status)

As has already been noted, the gender parity in terms of technology inclusiveness of the youth improves with higher income deciles. Therefore, inevitably poverty is an important indicator of technological inclusivity of the youth, particularly the female youth.

From Figure 10, three particularly important aspects can be identified. First, the share of female youths’ ownership of a mobile phone among poor households remains almost the same (around 34%–37%) across the age groups. It is mostly the females from the poor households who remain outside the reach of technological inclusivity. Second, the percentage of mobile ownership among the males is almost similar across the poor and non-poor categories. However, one aspect that this data cannot address is the type of mobile phones these two groups own (such as 2G/3G-enabled devices, smartphones, etc.). Lastly, for all three age categories, irrespective of poverty status, young females’ average mobile ownership is significantly lower than the average mobile ownership of males. The gendered divide in technological inclusivity is evident from this pattern.

Empirical Results

Results from the Household-level Decomposition

As has already been mentioned, for the household-level decomposition, the outcome variable of interest is the per capita mobile ownership of youth. Since the outcome variable is continuous, the simple linear regression technique has been used for the decomposition where the indicator variable is gender. The different socio-economic characteristics of the households have been used as the explanatory variable, as shown in Table 2. The results of the threefold decomposition are presented in Table 3, where prediction 1 is for male and prediction 2 is for female.

Table 2.

Variable Description.

Variables for Household-level Decomposition	Description
Rural	= 1 if the household is in a rural area; 0 otherwise
Migrant HH dummy	= 1 if any member of the household lives abroad; 0 otherwise
HH size	Number of members in the household
Age HH head	The age of the household head
Prop of women (age >15)	Number of women in the household aged over 15 divided by the household size
Male head	= 1 if the head of the household is male; 0 otherwise
HH total land pc	Household per capita total land
Middle income	= 1 if the household income is in the second quartile; 0 otherwise
Upper-middle income	= 1 if the household income falls in the third quartile; 0 otherwise
High income	= 1 if the household income falls in the fourth quartile; 0 otherwise
HH average education	Total education of the household member divided by household size
Barisal	= 1 if the household is in Barisal division; 0 otherwise
Chattogram	= 1 if the household is in Chattogram division; 0 otherwise
Rangpur	= 1 if the household is in Rangpur division; 0 otherwise
Khulna	= 1 if the household is in Khulna division; 0 otherwise
Mymensingh	= 1 if the household is in Mymensingh division; 0 otherwise
Rajshahi	= 1 if the household is in Rajshahi division; 0 otherwise
Sylhet	= 1 if the household is in Sylhet division; 0 otherwise
Variables used only for the individual-level decomposition in addition to the above-mentioned variables
Secondary education	= 1 if the youth has secondary education; 0 otherwise
High secondary education	= 1 if the youth has high secondary education; 0 otherwise
University education	= 1 if the youth has university education; 0 otherwise
Age youth	Age of the youth
Earner youth	= 1 if the youth is an earner; 0 otherwise
Married youth	= 1 if the youth is married; 0 otherwise

Source: The authors.

Table 3.

Blinder–Oaxaca Decomposition for the Household Per Capita Mobile Ownership among Youth.

Variables	(1)	(2)	(3)	(4)
Variables	Differential	Endowments	Coefficients	Interaction
Rural		–0.001***	0.046***	0.001***
		(0.000)	(0.006)	(0.000)
Migrant HH dummy		–0.003***	–0.018***	0.003***
		(0.001)	(0.002)	(0.001)
HH size		–0.000**	0.002	0.000
		(0.000)	(0.014)	(0.000)
Age HH head		–0.004***	0.003	0.000
		(0.001)	(0.014)	(0.001)
Prop of women (age >15)		0.002	0.164***	–0.019***
		(0.001)	(0.014)	(0.002)
Male head		–0.001**	0.129***	0.001***
		(0.000)	(0.014)	(0.000)
HH tot land pc		0.000	–0.000	–0.000
		(0.000)	(0.001)	(0.000)
Middle income		–0.000	–0.003	0.000
		(0.000)	(0.003)	(0.000)
Upper-middle income		0.002***	–0.013***	–0.001**
		(0.000)	(0.003)	(0.000)
High income		0.003***	–0.025***	–0.001***
		(0.001)	(0.003)	(0.000)
HH average education		0.012***	–0.120***	–0.010***
		(0.001)	(0.008)	(0.001)
Barisal		0.000	–0.001	–0.000
		(0.000)	(0.002)	(0.000)
Chattogram		–0.000	–0.006**	0.000*
		(0.000)	(0.003)	(0.000)
Khulna		–0.000	0.007***	0.000
		(0.000)	(0.002)	(0.000)
Mymensingh		–0.000	0.003***	0.000
		(0.000)	(0.001)	(0.000)
Rajshahi		–0.000	0.010***	0.000
		(0.000)	(0.002)	(0.000)
Rangpur		–0.001***	0.010***	0.001***
		(0.000)	(0.002)	(0.000)
Sylhet		0.000	0.007***	–0.000
		(0.000)	(0.001)	(0.000)
Total		0.007***	0.337***	–0.023***
		(0.002)	(0.004)	(0.003)
Prediction_1	0.794***
	(0.003)
Prediction_2	0.474***
	(0.003)
Difference	0.321***
	(0.004)
Constant			0.142***
			(0.029)
Observations	40,172	40,172	40,172	40,172

Source: The authors.

Note: Standard errors in parentheses.

***p < 0.01, **p < 0.05 and *p < 0.1.

The decomposition results show that the technological differential between males and females is 0.321, or 32 percentage points (Table 3). From the decomposition of this gender gap we find that the explained portion of the endowment effect is smaller in magnitude compared to the discriminatory part (the sum of the coefficient term and interaction term). Out of this total gap of 32.1 percentage points, only 0.7 percentage points are explained by the difference in observed characteristics and 31.4 percentage points (the sum of coefficient effect and interaction term) by the discriminatory behaviour against female youth. More explicitly, only 3% of the observed gender gap (of 32.1 percentage points) can be attributed to the observed characteristics, and 97% of this gap can be attributed to gender discrimination against female youths. This implies that if both males and females are treated in the same way (no difference in coefficient), then on average the per capita mobile ownership of female youths will increase by 31 percentage points.

On the other hand, the endowment effect of 0.007 implies that if males and females have the same observable characteristics, then the per capita mobile ownership of the females will increase by 0.007, that is, even if the males and females have the same characteristics, the average mobile ownership of the females will be lower compared to that of the males by 31 percentage points just due to the discriminatory behaviour. The coefficient term is large in magnitude and highly statistically significant.

The contribution of different variables to total endowments and the coefficient and interaction terms is also shown in Table 3. The positive sign of the coefficient of rural dummy implies that rural households show higher discriminatory behaviour against females compared to urban ones. Out of the total coefficient effect of 0.338, the contribution of the rural dummy is 0.049 (14%).

As the result suggests, having a migrant member in the household may contribute to reducing discrimination towards females by 1.9 percentage points.

Household financial status contributes to reducing the gender gap. Based on the household income, the households are clustered in four quartiles; the households in the bottom 25% are considered low-income households, the households in the next 25% are considered lower-middle-income ones, and in a similar fashion, the households in the top 25% percentile are considered high-income households. Compared to the low-income households, the middle-income, upper-middle-income and high-income households have lower gender difference due to discriminatory behaviour as reflected by the negative sign of the coefficient of these dummy variables.

The average education of the household members also contributes to reducing the discriminatory behaviour in the household against females. The results show that female youths face lower discrimination in a household with a high number of average years of education among the household members. The sex of the household head also appears to influence gender discrimination against females. If the household head is a male, the gender discrimination against females increases by 13 percentage points.

Among other factors, regional disparity also plays a role. Compared to that in Dhaka, gender discrimination in mobile ownership is significantly higher in Rajshahi and Sylhet. The discrimination against females is lower in Chattogram and Barisal compared to that in Dhaka.

Results from the Individual-level Decomposition

For the individual-level decomposition, the outcome variable is binary, and therefore Equations (1) and (2) cannot be estimated using the linear regression technique. In this study, the selection equation (mobile ownership) is estimated with the probit model where the dependent variable ‘mobile’ takes the value of 1 if the youth owns a mobile and 0 if the youth does not own a mobile. The marginal effects, along with the original coefficient estimates, are presented in Table 4. The results from the analysis show that the age of the youth has a positive and statistically significant impact on the probability of owning a mobile for both males and females (Table 4). Being an earner has a more substantial effect on the likelihood of owning a mobile for males compared to that for females. For male youths, being an earner increases the probability of owning a mobile by 21 percentage points, while for female youths the figure is only 13 percentage points.

Table 4.

Probit Model Results for the Youth Mobile Ownership.

Dependent variable: Mobile ownership status of the youth (probit regression)
Variables	Male		Female
Variables	Coefficient	ME	Coefficient	ME
Rural	0.007	0.0157	–0.193***	–0.076
	(0.025)		(0.019)
Middle income	0.152***	0.0337	0.116***	0.0461
	(0.031)		(0.025)
Upper-middle Income	0.272***	0.0587	0.258***	0.1027
	(0.032)		(0.026)
Higher income	0.389***	0.0822	0.475***	0.1876
	(0.036)		(0.029)
HH total land pc	0.007	0.0016	0.002	0.00083
	(0.007)		(0.004)
Secondary education (youth)	0.178***	0.0397	0.199***	0.0792
	(0.029)		(0.023)
High secondary education	0.800***	0.15	0.639***	0.2493
	(0.038)		(0.031)
University education	1.152***	0.144	1.104***	0.3832
	(0.106)		(0.075)
Barisal	–0.023	–0.032	–0.641***	–0.0680
	(0.029)		(0.038)
Chattogram	–0.110***	–0.026	0.056**	0.0224
	(0.032)		(0.026)
Khulna	0.054	0.0122	–0.141***	–0.0559
	(0.036)		(0.028)
Mymensingh	–0.016	–0.0038	–0.152***	–0.0598
	(0.051)		(0.041)
Rajshahi	0.122***	0.0269	–0.199***	–0.0782
	(0.039)		(0.030)
Rangpur	0.047	0.0105	–0.207***	–0.0815
	(0.037)		(0.031)
Sylhet	–0.259***	–0.0664	–0.424***	–0.1622
	(0.041)		(0.036)
Prop of women (age > 15)	0.114	0.0264	0.783***	0.3114
	(0.098)		(0.068)
Age youth	0.095***	0.0220	0.071***	0.0282
	(0.004)		(0.003)
Migrant HH dummy	0.071*	0.0159	0.435***	0.1712
	(0.038)		(0.029)
HH average education	0.017***	0.0038	0.059***	0.0235
	(0.006)		(0.005)
Earner youth	0.849***	0.208	0.327***	0.1294
	(0.029)		(0.031)
Married youth	–0.006	–0.00145	0.492***	0.1916
	(0.038)		(0.025)
Constant	–2.125***		–2.889***
	(0.081)		(0.072)
Observations	22,077		24,740

Source: The authors.

Note: Standard errors in parentheses.

***p < 0.01, **p < 0.05 and *p < 0.1.

The marginal effects coefficient estimates of the educational-attainment dummies of youth are found to be positive and statistically significant for all categories. Females with secondary education have a probability of owning a mobile about 7.9 percentage points higher than female youth having primary education or no education (the base category). This probability increases to 24.9 percentage points if the individual attains higher secondary education and rises further to 38.3 percentage points if a university or higher degree is attained. For male youths, the corresponding probabilities are 3.9, 15 and 14.4 percentage points higher, respectively.

The migration status of a household has a strong positive effect on the probability of owning a mobile phone among the female youths. A female youth belonging to a household having one or more members living abroad has a higher probability (by 17 percentage points) of owning a mobile.

The impact of household wealth is captured by the dummy variables representing the different income categories of the household to which a youth belongs. Here, a youth living in the poorest quartile has been used as the base category. The result shows, whether it is male or female, youth from richer households have a higher probability of owning a mobile phone.

Decomposition Results

The B–O decomposition results following Fairlie (1999) for individual mobile ownership are presented in Table 5. The results show that the mobile ownership rate among the male youth is 79%, while it is only 47% among the female youth. The overall gender difference in mobile ownership is 0.32, or 32 percentage points, of which only 24% (0.078) is explained by the differences in observed characteristics, and the rest due to discriminatory behaviour.

Table 5.

Blinder–Oaxaca Decomposition Results Following Fairlie’s Technique (1999).

Male youth mobile ownership rate	0.79
Female youth mobile ownership rate	0.47
Male–female gap	0.32
Contribution of gender difference
Rural	0.000 (0.000)
Middle income	0.000* (0.000)
Upper-middle income	0.002*** (0.000)
Higher income	0.005*** (0.000)
HH tot land pc	0.000 (0.000)
Secondary education youth	–0.004*** (0.001)
High secondary education youth	0.031*** (0.001)
University education youth	0.002*** (0.000)
Barisal	0.000 (0.000)
Chattogram	0.000 (0.000)
Khulna	0.000 (0.000)
Mymensingh	0.000 (0.000)
Rajshahi	0.000** (0.000)
Rangpur	–0.000 (0.000)
Sylhet	0.001*** (0.000)
Prop of women (age > 15)	–0.002 (0.001)
Age youth	–0.037*** (0.001)
Migrant HH dummy	–0.000 (0.000)
HH average education	0.004*** (0.001)
Earner youth	0.074*** (0.003)
Married youth	0.001 (0.005)
Total explained	0.078 (24%)
Total unexplained	0.242 (76%)

Source: The authors.

Note: Standard errors in parentheses.

***p < 0.01, **p < 0.05 and *p < 0.1.

Among the explanatory variables, the earning status of the youth has the highest and most significant contribution in the explained part of the gender differences. Holding other variables constant, if the earning status of a female is replaced by that of a male, the probability of owning a mobile will rise significantly for the female youth. Lower levels of education among the female youth also contribute to the gender differences in mobile ownership. The income status also has a significant contribution in explaining the gender differences in mobile ownership, but the group differences in regional distribution make no or little contribution to the gender gap in mobile ownership. The contribution of household income ranges from 0.2 percentage points to 0.5 percentage points, and the contribution of youth education ranges from 0.2 percentage points to 3.1 percentage points.

Based on the analyses presented in this section, it can be inferred that the gender gap in technological inclusiveness (measured in terms of mobile ownership) observed from descriptive statistics is largely due to gender discrimination against females. It is also evident from the empirical analysis at both the household level and individual level that the education of the youth and the average years of schooling of the household members play essential roles in reducing such gender discrimination. The sex of the household head also matters in the case of technological inclusiveness of females. Among other factors, the income level of households, the migration status of households, etc., contribute to improving the overall gender parity in terms of mobile ownership among young females.

Conclusion

Achieving the SDGs in the era of FIR would not be possible if the dividends of technology are not equally shared by men and women, the rich and poor and rural and urban people. Ensuring such equity demands equipping the youth with modern ICT. In this respect, it is important to identify the key barriers to the technological inclusivity of the youth and to analyse the gendered digital divide.

To identify the key barriers to the technological inclusivity of youth and to confirm the gender discrimination against females, this study has employed standard decomposition techniques. The nature of the digital divide in access to technology (mobile phones) by region, income group and gender has been observed in detail. The findings of the study show that households from richer income deciles have more technological inclusivity than households from poorer income deciles. Youths from regions with higher economic activities have more access to technologies than those from lagging regions. Moreover, in all the cases, female youths’ inclusiveness with regard to technology has been found to be significantly lower than the male youth’s inclusiveness to technology.

The econometric analysis shows that a significant part of the gender differences observed in terms of mobile ownership originates from discriminatory behaviour against female youth. Relating it to findings from previous studies, one of the primal causes of such gender discrimination against females could be attributed to social stigma. It is also observed in the study that the factors such as education of the youth, the average years of schooling of the household members, the income level of households, the migration status of the households, etc., contribute in improving the overall gender parity in terms of mobile ownership among young females. Based on the findings and observations, this study proposes the following set of recommendations for ensuring higher youth inclusivity in technology and addressing the gendered digital divide in Bangladesh.

A revamped focus is needed in formulating a gender-inclusive ICT policy and action plan: There is a need for revamped action for formulating a gender-inclusive ICT policy and action plan. In this regard, GoB should take thorough stock of the present scenario of gender inclusiveness in technology. Based on the analysis, a short- to medium-term plan should be implemented by the GoB. Also, the ‘Digital Bangladesh’ agenda should be reinvigorated.

An ICT-friendly fiscal policy should be considered with priority: As women are lagging behind, some special packages can be offered to them, such as offering a tax rebate on purchasing a technological device (such as a computer or mobile phone) or offering internet services at subsidized prices. Female students can be made eligible for VAT (value-added tax) refund while purchasing computers or mobile phones. If required, the government can think of tax rebates or reductions in the existing tariffs for the telecommunications industry while importing ICT equipment.

Awareness must be created among people regarding the benefits of technology in eliminating social stigma: The social stigma around the usage of technological devices by female youths has been identified as one of the stumbling blocks for the technological inclusiveness of females. The ICT division of the government should undertake aggressive social campaigns through print and electronic media, as well as social media, to raise social awareness on the benefits of technology.

There must be a special focus on the marginalized groups of people and lagging regions: The policy needs to focus on people from lagging regions and disadvantaged populations. In addition to rural-centric policies, emphasis should be laid on people from ethnic minorities and marginalized communities (such as Bede, Dalits, etc.). The policy should consider access to ICT as a basic right rather than a privilege.

Footnotes

Declaration of Conflicting Interests

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Notes

References

Ben

(2008). A Stata implementation of the Blinder–Oaxaca decomposition. Stata Journal , 8(4), 453–479.

Blinder

(1973). Wage discrimination: Reduced form and structural estimates. The Journal of Human Resources , 8(4), 436–455.

Castells

(2004). Informationalism, networks and the network. In Castells

(Ed.), The network society: A cross-cultural perspective (pp. 3–45). Edward Elgar.

Chakraborty

Bosman

M. M.

(2005). Measuring the digital divide in the United States: Race, income, and personal computer ownership. The Professional Geographer , 57(3), 395–410.

Chinn

M. D.

Fairlie

R. W.

(2004). The determinants of the global digital divide: A cross-country analysis of computer and internet penetration (NBER Working Paper No. 10686). Economic Growth Center, Yale University.

Cooper

(2006). The digital divide: The special case of gender. Journal of Computer Assisted Learning , 22(5), 320–334.

Crewe

Harrison

(1998). Whose development? An ethnography of aid . Zed Books.

Fairlie

R. W.

(1999). The absence of the African–American owned business: An analysis of the dynamics of self-employment. Journal of Labor Economics , 17(1), 80–108.

Fairlie

R. W.

(2002). Race and the digital divide (Santa Cruz Working Paper). University of California.

10.

Fairlie

R. W.

(2005). An extension of the Blinder–Oaxaca decomposition technique to logit and probit models. Journal of Economic and Social Measurement , 30(5), 305–316.

11.

Fuchs

Horak

(2008). Africa and the digital divide. Telematics and Informatics , 25(2), 99–116.

12.

HIES. (2016). Bangladesh Bureau of Statistics, Statistics Division, Ministry of Planning.

13.

Hoque

S. M.

Mahiuddin

Alam

S. M.

(2011). E-Governance: A way forward to digital Bangladesh. Journal of the Department of Government and Politics, Jahangirnagar University , 30, 127–136.

14.

Islam

M. A.

Tsuji

(2011). Bridging digital divide in Bangladesh: Study on com- munity information centers. The Electronic Library , 29(4), 506–522.

15.

Jones

Johnson-Yale

Millermaier

Pérez

F. S.

(2009). U.S. college stu- dents’ internet use: Race, gender and digital divides. Journal of Computer-Mediated Communication , 14(2), 244–264.

16.

Karim

M. A.

(2010). Digital Bangladesh for good governance . Bangladesh Development Forum.

17.

Keegan

M. E.

(2004). Digital divide in computer access and use between poor and non- poor youth. The Journal of Sociology and Social Welfare , 31, 91.

18.

Korupp

E. S.

Szydlik

(2005). Causes and trends of the digital divide. European Sociological Review , 21(4), 409–422.

19.

Livingstone

Helsper

(2007). Gradations in digital inclusion: Children, young people and the digital divide. New Media and Society , 9(4), 671–696.

20.

Oaxaca

(1973). Male–female wage differentials in urban labour markets. International Economic Review , 114(3), 693–709.

21.

Oaxaca

Ransom

M. R.

(1994). On discrimination and the decomposition of wage differentials. Journal of Econometrics , 61(1), 5–21.

22.

Prieger

J. E.

W. M.

(2008). The broadband digital divide and the nexus of race, competition, and quality. Information Economics and Policy , 20(2), 150–167.

23.

Rice

R. E.

Katz

J. E.

(2002). Comparing internet and mobile phone digital divides. Proceedings of the American Society for Information Science and Technology , 39(1), 92–98.

24.

Schwab

(2016). The fourth industrial revolution: What it means, how to respond . World Economic Forum. https://www.weforum.org/agenda/2016/01/the-fourth-industrial-revolution-what-it-means-and-how-to-respond/

25.

Sinning

Hahn

Bauer

T. K.

(2008). The Blinder–Oaxaca decomposition for nonlinear regression models. The Stata Journal , 8(4), 480–492.

26.

Sweetman

(1998). Gender and technology (Vol. 7, No. 2). Oxfam GB.

27.

Tran

M. C.

Labrique

A. B.

Mehra

Ali

Shaikh

Mitra

Christian

West

Jr (2015). Analyzing the mobile ‘digital divide’: Changing determinants of household phone ownership over time in rural Bangladesh. JMIR mHealth uHealth , 3(1), e24.

28.

Venkatesh

Morris

M. G.

Ackerman

P. L.

(2000). A longitudinal field investigation of gender differences in individual technology adoption decision-making processes. Organizational Behavior and Human Decision Processes , 83(1), 33–60.

29.

Warschauer

(2003). Economy, society, and technology: Analyzing the shifting ter- rains. In Warschauer

(Ed.), Technology and social inclusion (p. 12). The MIT Press.

30.

Zhou

Singh

Kaushik

P. D.

(2011). The digital divide in rural South Asia: Surveyevidence from Bangladesh, Nepal and Sri Lanka. IIMB Management Review , 23(1), 15–29.