Internet use, spatial variation and its determinants among reproductive age women in Ethiopia

Abstract

Background

The Internet is the preferred source of health information for retrieving relevant information. In Ethiopia, the Internet penetration rate is improving year to year, but it is still at a low level compared to the rest of the world and neighboring African countries. Due to a lack of adequate information, it is important to assess Internet use, spatial variation, and determinants of Internet use among reproductive-age group women in Ethiopia.

Method

Secondary data from EDHS 2016 were used to analyze 15,683 women aged 15–49 years. Spatial analysis was performed using ArcGIS 10.7. The Bernoulli model was used by applying Kuldorff’s methods using SaTScan 10.1.2 software to analyze the purely spatial clusters of Internet use. A multilevel mixed-effect logistic regression was applied to estimate community variance to identify individual- and community-level factors associated with Internet use. All models were fitted in STATA version 17.0, and finally, the adjusted odds ratio (AOR) with a corresponding 95% confidence interval (CI) was reported.

Result

The magnitude of Internet use was 4.97% ± 95% CI (4.63–5.32). The overall average age of women was 24.21 ± 8.06 years, with the age range 15–24 years constituting the larger group (39.2%). Women with secondary and above education [AOR = 6.47; 95% CI (5.04, 8.31)], unmarried [AOR = 2.60; 95% CI (1.89, 3.56)], rich [AOR = 1.95; 95% CI (1.00, 3.80)], own a mobile phone [AOR = 3.74; 95% CI (2.75, 5.09)], media exposure [AOR = 2.63; 95% CI (2.03, 3.42)], and urban [AOR = 1.80; 95% CI (1.08, 3.01)] had higher odds of Internet use. The spatial variation in Internet use was found to be nonrandom (global Moran’s I = 0.58, p value < 0.001). Fifty-seven primary clusters were identified that were located in Addis Ababa city with a relative likelihood of 10.24 and a log-likelihood ratio of 425.16.

Conclusions

Internet use among reproductive-age women in Ethiopia is 4.97 and has significant spatial variation across the country. Both community- and individual-level factors affect Internet use in Ethiopia. Therefore, educating women, improving access to media, encouraging women to use family planning, and supporting household wealth could improve women’s Internet use.

Background

The Internet is a world-wide public computer network. The Internet provides a global platform connecting thousands of networks around the world. Currently, the Internet is an integral part of the daily life of most people [1]. Internet use is the use of the Internet by individuals whether at home, in their organizations, or at their institutions [2]. The development of technology, especially the Internet, can change all aspects of life and has a variety of information from all corners of the world without any restrictions. The Internet has expanded as a resource for knowledge and access to health care information, as well as for social and economic opportunities [3, 4].

Today, an estimated 5.3 billion of the world’s population use the Internet [5]. According to the Internet World Statistics report in 2022, the Internet penetration rate was globally 64.2%, and it will pick in Western countries at approximately 92%. The Internet penetration rate in Africa was 43.0% and in Ethiopia 17.7% [6]. Internet-based communication for health purposes via the Internet is growing and has a significant role in the health system. It has been implemented for various types of programs across different levels of the health sector and health policy. Integrating Internet-based communication in a health program has contributed much to the success of the program [7, 8].

In low-income countries, half of women have no Internet access, and there is a 23% gap in mobile Internet access across low- and middle-income countries. This reflects that women are disproportionately missing out on the well-being benefits of Internet use compared to men [9].

The Internet penetration rate was different among countries. In Sweden, 84% [10]; in Italy, 95%; in Canada, 93.50%; in China, 88.7% [11, 12]; in Poland, 66.7% [13]; and in sub-Saharan Africa, improvement from 2020 (2.1%) to 2022 (34%) [14]. According to a 2022 report on Internet users in Africa by country, the least Internet usage among women was found in the Central African Republic (7.1%), Eritrea (8%), Comoros (8.5%), and South Sudan (10.9%) [15]. Most African countries are the least developed countries and have lower socioeconomic development and low access to technology-related infrastructures in these countries [16]. Based on International Technology Union (ITU) data at the start of 2023, the Internet penetration rate in Ethiopia was estimated at 16.7%. Female Internet use as a percentage of the total female population is estimated at 14% [17, 18].

Several factors that have an impact on Internet use were identified in different parts of the world, including sub-Saharan Africa. These are age [3, 19,20,21], education level [22,23,24,25], marital status [26,27,28], wealth index [22, 29, 30], having a mobile phone [9, 20, 31], use of family planning [32,33,34,35], place of residence [30, 36] and media exposure, which are significantly associated with Internet use.

A study conducted in Kuwait showed that 25.6% of women in the reproductive-age group received health-related information through the Internet. According to this study, except for age and marital status, all sociodemographic factors affected looking for health-related information through the Internet [37].

In developing countries such as Iran, a lack of regular Internet access limits access to some health services, such as general counseling and finding information about medications [22]. In a study conducted in Turkey, women had greater information needs during pregnancy. The information gathered during this time has an impact on labor, delivery, and the postpartum period [38]. In Ethiopia’s reproductive-age, women’s exposure to the Internet for searching information about family planning is 12%, which is very low [7].

Internet use to access health information leads women to have opportunities to improve their health and make health-related decisions. However, there is a lack of adequate information about Internet use status, spatial variation, and its associated factors, particularly among reproductive-age women searching for health information in Ethiopia.

Therefore, this study aimed to determine Internet use, spatial variation, and factors associated with Internet use among reproductive-age women in Ethiopia by using data from EDHS 2016. This study may differ from previous studies because it used nationally representative data. The finding will be important to have evidence-based interventions to enhance the Internet use habits of women for accessing updated health information and serve as literature for further studies in the field.

Methods

Study setting, design and period

The study was conducted in Ethiopia. Ethiopia is the second most populous country located in the Horn of Africa (3^o-14^o N, 33^o – 48°E). The current population of Ethiopia is 123,136,977 as of May 18, 2023, with 78.7% living in rural areas based on Worldometer elaboration of the latest United Nations data [39]. Currently, in Ethiopia, there are 11 regional states (Afar, Amhara, Benishangul-Gumuz, Gambela, Harari, Oromia, Sidama, South West Ethiopia Peoples’, Somali, SNNP, and Tigray) and 2 administrative cities (Addis Ababa and Dire-Dawa). The 2016 EDHS data were collected only from 9 regional states, and two cities administrative. Therefore, this study also focuses on only nine regions and two city administrations (Fig. 1). This study used secondary data from EDHS 2016. In EDHS 2016, the Central Statistical Agency (CSA) conducted a community-based cross-sectional study from January 18 to June 27, 2016 [40]. The women’s data (IR) from the 2016 EDHS were used.

Fig. 1

Study area map (Ethiopia) 2016. Shapefile source: CSA, 2013; URL: https://africaopendata.org/dataset/ethiopia-shapefiles

Source and study population

All reproductive-age women living in Ethiopia prior to the survey period were the source population, whereas all reproductive-age women who were located in selected EA in Ethiopia during the study period were the study population. All reproductive-age women who were usually members of the selected households or who spent the night before the survey in the selected households were included in the study. Areas with missing geographic coordinates for spatial analysis were excluded. The data were obtained from the DHS Program through an online request by explaining its objective and significance (www.dhsprogram.com). From 16,583 eligible women for individual interviews (IR), 15,683 women, with a response rate of 95%, completed the interviews [40].

Sample size and sampling procedure

A total of 18,008 households were selected for the sample, of which 17,067 were occupied. Of the occupied households, 16,650 were successfully interviewed, yielding a response rate of 98%. In the interviewed households, 16,583 eligible women were identified for individual interviews (IR). Finally, interviews were completed with 15,683 women, yielding a response rate of 95% [40].

EDHS 2016 used the 2007 Population and Housing Census as a sampling frame for enumeration areas (EAs) in Ethiopia. The 2007 census contained a complete list of 84,915 enumeration areas (EAs), and one EA is a geographical area covering 181 households on average. A two-stage stratified sampling technique was employed to select representative samples for the country as a whole. The regions in the country were stratified into urban and rural areas, resulting in 21 sampling strata. Then, samples of enumeration areas (EAs) were selected in each stratum in two stages.

In the first stage, 645 EAs (202 in urban and 443 in rural areas) were selected with a probability proportional to the EA size. The EA size is the number of residential households in the EA as determined in the 2007 Ethiopian Population and Housing Census. In the second stage, a fixed number of 28 households per cluster were selected with an equal probability of systematic selection from the newly created household listing. All women aged 15–49 years who were usually members of the selected households or who spent the night before the survey in the selected households were eligible for the female survey [40].

For this study, individual record (IR) datasets were used. A total of 15,683 women aged 15 to 49 years in the 2016 EDHS with complete responses to all variables of interest were selected for analysis. Due to the nonproportional allocation of the sample to different regions and their urban and rural areas and the possible differences in response rates, a sampling weight was used in all analyses using the 2016 EDHS data to ensure the actual representativeness of the survey results at both the national and domain levels.

Data collection tool and procedures

EDHS 2016 used five questionnaires. These are the household questionnaire, the woman’s questionnaire, the man’s questionnaire, the biomarker questionnaire, and the health facility questionnaire. These questionnaires, based on the DHS Program’s standard DHS questionnaires, were adapted to reflect the population and health issues relevant to Ethiopia. After all, the questionnaires were finalized in English and translated into Amarigna, Tigrigna, and Oromiffa.

The woman’s questionnaire was used to collect information from all eligible women aged 15–49. These women were asked questions on their background characteristics (including age, education, and media exposure), family planning (including knowledge, use, and sources of contraceptive methods), ANC (delivery and postnatal care), and others [40].

CSA recruited and trained 294 people for the main fieldwork to serve as team supervisors, field editors, interviewers, secondary editors, and reserve interviewers. The training was given from December 14, 2015, to January 17, 2016, at the Debre Zeit Management Institute in Bishoftu. The training was focused on instruction regarding interviewing techniques and field procedures, a detailed review of questionnaire content, instruction on how to administer the paper and electronic questionnaires, mock interviews between participants in the classroom, and practice interviews with real respondents in areas outside the survey sample [40].

For this study the data were obtained from the DHS Program through an online request by explaining its objective and significance (www.dhsprogram.com). After permission was approved by the DHS measuring program to download the datasets, the EDHS 2016 dataset and GPS dataset (zip file) were downloaded, and the outcome and the predictor variables were extracted from the women (IR) dataset using STATA.

Study variables

Dependent variable

The outcome variable of this study was Internet use, which is taken as binary (“1” use the Internet or “0” not use the Internet) based on whether the women use the Internet in the last 12 months or not. If the responses of ‘never and yes but don’t know when’ are taken as not used Internet (No), else used Internet (Yes) [41].

Independent variable

Individual-level factors: maternal age, marital status, highest education level, occupation, household wealth index, household electricity, having own mobile phone, ever used family planning, pregnancy, and media exposure.

Community-level factors: Place of residence, region, community-level media exposure, community maternal education, and community-level poverty.

Operational definitions

Internet use is the variable generated from the EDHS dataset, which was assessed by asking “In the last 12 months, have you used the Internet?”. The possible responses were “never, yes in the last 12 months, yes before the last 12 months’ and yes but don’t know when”. The responses of ‘never and yes but don’t know when’ are taken as not used Internet (No), else used Internet(Yes)( [42]).

The respondent’s occupation was assessed by asking women about their current occupation or occupation in the last 12 months. Then, recategorized as not working and working (other several working classes into working). However, just because women are “not working” doesn’t mean they can’t do any work; rather, they are restricted to activities at home.

Exposure to media: data management and analysis

Women’s exposure to mass media is defined as women aged 15–49 who report exposure to either radio, television, or newspapers at least once a week. EDHS collected by asking “How often do you read a newspaper, listen to the radio, and watch television (TV) at least once a week, less than once a week, or not at all?”. These variables were first recoded into “yes” and “no”. Not at all, and less than once a week as “no” and else “yes”. Then exposure to media variable was considered as “exposed” if the respondent was exposed to at least one of the three media ( > = 33%) and otherwise “non-exposed” ( [43]).

Except for region and place of residence, other community-level explanatory variables were created by aggregating individual-level characteristics at the community (cluster) level. Aggregated variables are categorized as low and high based on the proportion values distribution that is calculated for each community. The histogram was used to check the proportion value distribution. If the distribution is a normal mean value, else median value was used as a cut-off point for the categorization.

Community-level media exposure was created from individual-level women’s exposure to mass media. By computing the proportion of women exposed to at least one of these media, then were categorized into low (if < 0.20) and high (if ≥ 0.20) community-level media exposure. Since the data is not normally distributed a national median value was used as the cutoff point.

Community poverty was obtained by recording the poorer and poorest as poor from the household wealth index. Then, the proportion of poor household wealth index was calculated and categorized as high poverty level (those with > = 0.26) and low poverty level (those with < 0.26) using the national median value.

Community-women education was obtained by aggregating the individual level of woman’s education to the cluster level by taking the proportion of women who attended secondary and above education. Then categorized low (those with < 0.125) and high (those with ≥ 0.125) levels of community women’s education using a national median value.

Data management and analysis

STATA 17, ArcGIS 10.7, and SatScan 10.1.2 software was used to perform data analysis. Before any statistical analysis, using STATA, the data were weighted by sampling weight (women sample weight) to assure the representativeness of the survey (to adjust the nonproportional allocation of samples to strata and regions) and to tell the software to consider the sampling design when calculating standard errors to obtain reliable statistical estimates.

For spatial analysis, the cleaned data were cross-tabulated to the outcome variable with cluster and then exported to Microsoft Excel. The spatial data that contain the latitude and longitude zero (0) coordinates were removed using ArcGIS 10.7. Then, Excel data were combined with GPS data for spatial analysis. Descriptive statistics and summary statistics were presented using text, tables, and figures.

Multilevel analysis

Before conducting any statistical analysis, the data were weighted using sampling weight to restore the survey’s representativeness and account for the sampling strategy. This was done to produce accurate statistical estimations. STATA 17 was used to perform descriptive and summarizing statistics. With the hierarchical nature of the EDHS data and the nesting of women inside clusters, it was reasonable to assume that women within a cluster may be more similar to one another than women throughout the country. This implies the need to use an advanced model to account for cluster heterogeneity. Therefore, a multilevel binary logistic regression model (multilevel mixed-effect regression) was applied.

Model comparison was performed based on the Akaike Information Criteria (AIC) and Deviance (-2LL). The model with the lowest information criterion and lowest deviance was considered the best-fitting model. The likelihood ratio test, intraclass correlation coefficient (ICC), median odds ratio (MOR), and proportional change in variance (PCV) were computed to measure the variation between clusters. ICC quantifies the degree of heterogeneity of Internet use between clusters (the proportion of the total observed individual variation in Internet use that is attributable to between cluster variations).

$\:ICC=\frac{\text{C}\text{V}}{(\text{C}\text{V}+\frac{{{\uppi\:}}^{2}}{3})}$, but MOR quantifies the variation or heterogeneity in outcomes between clusters and is defined as the median value of the odds ratio between the cluster at high Internet use and the cluster at lower use when randomly picking out two clusters (EAs).

$$\text{MOR}=\text{e}\text{x}\text{p}\left(0.6745*\sqrt{2*CV}\right) \sim \text{MOR}=\text{e}\text{x}\text{p}\left(0.95*\sqrt{CV}\right)$$

CV indicates cluster variance.

PCV measures the total variation attributed to individual and community-level factors in the multilevel model compared to the null model [44].

$$$$\text{PCV}=\frac{\text{Var}(\text{null model})-\text{Var}(\text{full model})}{\text{Var}(\text{null model})}$$$$

Due to the hierarchical nature of the data, multilevel mixed-effect logistic regression was applied at the individual and community (cluster) levels to identify factors associated with Internet use for health information. Four models were built using multilevel logistic regression analysis. The first model was an empty (null) model to determine the degree of cluster variability in Internet use, without any explanatory variables. A second model (identified associations between individual-level factors and Internet use) and a third model (identified associations between community-level variables and Internet use) were used. A fourth model (individual- and community-level models) was fitted simultaneously with individual- and community-level variables. The multilevel mixed-effect regression p value < 0.05 and AOR with a 95% CI (confidence interval) were used to identify significant predictors of Internet usage.

Spatial analysis

Spatial autocorrelation analysis

The spatial autocorrelation (Global Moran’s I) statistic measure was used to determine whether Internet use among reproductive age group women for health information is dispersed, clustered, or randomly distributed in Ethiopia. Moran’s I is a spatial statistic used to measure autocorrelation in space by taking the entire dataset and generating a single output value ranging from − 1 to + 1. Moran’s, I value close to − 1 shows dispersed Internet use, close to + 1 shows clustered Internet use, and Moran’s I value zero shows random distribution. A statistically significant Moran’s I (p < 0.05) indicates that the spatial distribution of Internet use is nonrandom and suggests the existence of spatial autocorrelation. This leads to the rejection of the null hypothesis [45].

Getis-Ord Gi ^* hotspot analysis

Hotspot analysis (the Getis-Ord Gi^* statistic) was computed to measure how spatial autocorrelation differs through study location by calculating Gi* statistics for each area. The z score was calculated to ensure the statistical significance of clustering at a p value < 0.05 with a 95% CI. If the Z score is between − 1.96 and + 1.96, the p value would be larger than 0.05, and the null hypothesis cannot be rejected, which implies that there was a random pattern. If the z score falls outside the range, the observed spatial pattern is significant, and the p value would be small; thus, the null hypothesis will be rejected. If the z score is less than − 1.96, the low value of Internet use is clustered (cold spot), and if it is greater than + 1.96, the high value of Internet use is clustered (hotspot) [46].

Spatial interpolations

The kriging spatial interpolation technique was applied to predict the penetration of the Internet in unsampled areas based on the values observed from sampled areas. There are various deterministic and geostatistical interpolation methods [47]. This study used the ordinary kriging spatial interpolation method since it had a smaller residual/lowest mean predicted error (MPE) and root mean square error, the best-fitting interpolation technique for Internet use. These small values ​​indicate that the predicted values ​​are close to the observed values ​​and vice versa [48].

Spatial scan statistics

Bernoulli-based model spatial Kuldorff’s Scan statistics were used to determine the community locations of the statistically significant spatial windows for Internet use using SatScan version 10.1 software [49]. A scan window moving through the study area with women using the Internet is used as a case, and women not using the Internet are used as controls to fit Bernoulli’s model. The default maximum size of spatial clusters < 50% of the population is used as the upper bound, and the minimum cases in clusters for high rates is 2. Small and large clusters are found, and clusters containing clusters exceeding the maximum boundary are ignored in a circular window. A selection of nonoverlapping options in SatScan version 10.1.2 was employed to generate primary and secondary clusters. A likelihood ratio test statistic and the p value were used to determine significant clusters for each possible cluster. The most likely performing cluster was the scanning window with a maximum likelihood. The primary and secondary clusters were established and ranked based on their likelihood test, and a p value less than 0.05 was considered statistically significant [50]. ArcGIS software version 10.7 was used to map the cluster and attribute of Internet use produced by SatScan 10.1.2.

Results

Background characteristics of the study participant

A total of 15,683 weighted samples of reproductive-age women with a 95% response rate were included in the study. Approximately 23 (3.6%) clusters (EA) were missing geographical coordinates (zero longitudes and latitude) in the spatial data.

Of the study participants, 6143 (39.2%) women were aged between 15 and 24 years, and the overall average age of respondents was 28.17 ± 9.16 (mean ± SD) years. Approximately 5,590 (35.0%) of the participants attained primary education. The majority of the participants 10,224 (65.2%) were married. Approximately 7263 (46.3%) of the respondents were from rich households, and 4283 (27.3%) had mobile phones. Of the total 4140 (26.4%) respondents exposed to mass media, approximately 660 (15.9%) used the Internet for health purposes (Table 1).

The majority of the study participants were rural dwellers 12,207 (77.8%). Most of the study participants were from the Oromia 5700 (36.3%), followed by the Amhara 3,714 (23.7%) and SNNP 3,288 (21.0%) regions. Regarding community media exposure and community women’s education, approximately 8,949 (57.1%) and 8315 (53.0%) were from communities with low media exposure and high women’s education, respectively (Table 2).

Table 1 Individual-level characteristics and cross tabulation with internet use among reproductive-age women in Ethiopia, 2016

Table 2 Community-level characteristics and cross tabulation with internet use among reproductive age women in Ethiopia, 2016

Proportion of internet use by regions in Ethiopia

The proportion of Internet use varied across regions. The highest percentage of Internet use was observed in the Addis Ababa 306 (39.3%), Oromia 160 (20.58%), and Amhara 122 (15.61%) regions. On the other hand, the lowest proportion of Internet use was found in the Afar (0.45%), Gambella (0.37%), and Benishangul (0.56%) regions (Fig. 2).

Fig. 2

Spatial variation in internet use across regions in Ethiopia, EDHS 2016

The proportion of internet use

In this study, a total of 779 (4.97%) ± (95% CI 4.87–5.59) women in the reproductive-age group used the Internet to obtain health information.

Multilevel analysis

In multilevel mixed-effect regression, a total of four models (null model, model with individual-level variables, model with community-level variables, and the final model that was model with both individual and community-level variables) were computed. The final model (model IV) was the best-fit model for the data since it had the lowest deviance (highest log likelihood) and lowest AIC value. The random intercepts and the fixed effects (a measure of association) for the use of the Internet are presented in Tables 3 and 4, respectively.

The random effect results in a multilevel analysis

The results of the null model (Model I) showed that there was statistically significant variability in the odds of Internet use between communities (CV = 5.13, p value = 0.000). The ICC value in the null model was 60.95%, indicating that 60.95% of the total variability in Internet use was attributable to the between-group variation, while the remaining 39.05% was explained by the between-individual variation. The median odds ratio also revealed that Internet usage among reproductive-age women was heterogeneous among clusters. The MOR was 8.61, indicating that if we randomly select two women from two different clusters, women in the cluster with higher Internet usage had 8.61 times higher odds of using the Internet than women in the cluster with lower Internet use (Table 3).

In Model II, only individual-level variables were added. The results showed that women’s age, highest education level, marital status, wealth index, own mobile phone, electricity, family planning, and exposure to mass media were significantly associated with Internet use. The ICC in Model II indicated that 20.46% of women’s Internet use variation was attributable to differences across communities. As shown by the PCV, 83.52% of the variance in Internet use across communities was explained by individual-level characteristics.

In Model III, only community-level variables were added, and the results revealed that except for community-level education, all variables were significantly associated with the use of the Internet. The ICC in Model III implied that differences between communities account for approximately 17.92% of the variation in women’s Internet usage. In addition, the PCV indicated that 86.01% of the variation in Internet use between communities was explained by community-level characteristics.

After the inclusion of both the individual- and community-level variables in model IV, the variation in the odds of Internet use between communities remained statistically significant (CV = 0.51, p value = 0.000). As shown by the estimated ICC, 13.46% of the variability in Internet use was attributable to differences between communities. The PCV indicated that 90.03% of the variation in Internet use across communities was explained by both individual- and community-level factors included in model IV (Table 3).

Table 3 Measures of variation & model fit statistics in internet use among reproductive age women in Ethiopia, 2016

The fixed effect results in a multilevel analysis

In the multivariable multilevel mixed-effect regression analysis, the final model (Model IV) included both the individual- and community-level characteristics simultaneously. Women’s age, women’s education, women’s marital status, wealth index, own mobile phone, family planning, exposure to mass media, and region were the significant determinants of Internet use.

The odds of women who attained secondary and above education were 6.47 times [AOR = 6.47; 95% CI (5.04, 8.31)] more likely to use the Internet than women who attained primary education. Single/unmarried women were 2.60 times [AOR = 2.60; 95% CI (1.89, 3.56)] more likely to use the Internet for health than married women. Women from rich and middle-sized households were 1.95 times [AOR = 1.95; 95% CI (1.00, 3.80)] and 2.39 times [AOR = 2.39; 95% CI (1.20, 4.77)] more likely to use the Internet for health purposes, respectively, than women from poor households.

Women with mobile phones were 3.74 times [AOR = 3.74; 95% CI (2.75, 5.09)] more likely to use the Internet than women who did not have mobile phones. The odds of women who ever used family planning were 1.38 times [AOR = 1.38; 95% CI (1.04, 1.85)] more likely to use the Internet than women who did not use family planning. The odds of using the Internet for health purposes among women who were exposed to mass media were 2.63 times [AOR = 2.63; 95% CI (2.03, 3.42)] higher compared to women who were not exposed to mass media (Table 4).

Among community-level variables, the odds of women in Addis Ababa were 2.60 times [AOR = 2.60; 95% CI (1.56, 4.35)], and those in Dire Dawa were 2.99 times [AOR = 2.99; 95% CI (1.27, 6.96)] more likely to use the Internet than women in the Tigray region. The odds of women in Somali were 2.63 times [AOR = 2.63; 95% CI (1.08, 6.41)] more likely to use the Internet than in the Tigray region. The odds of women in urban areas were 1.80 times [AOR = 1.80; 95% CI (1.08, 3.01)] more likely to use the Internet than those in rural residences (Table 4).

Table 4 Individual- and community-level factors associated with internet use among reproductive-age women in Ethiopia, 2016

Spatial analysis

Spatial autocorrelation

The findings showed that the distribution of Internet use among reproductive-age women was nonrandom in Ethiopia. The result is positive and skewed to the right. The global Moran’s I test was 0.58 with a significant p value (p < 0.001). As global Moran’s I was significant and greater than zero, it showed that the spatial variation of Internet use was clustered. This indicates that there is less than a 1% possibility of occurrence in random chance. Therefore, the null hypothesis was rejected at a Moran’s I test of 0.58. A null hypothesis stated that the use of the Internet was distributed randomly (Fig. 3).

Fig. 3

Spatial autocorrelation of internet use among reproductive-age women in Ethiopia, 2016

Getis Ord Gi* hotspot analysis

The Getis Ord Gi* statistical analysis identified the significant spatial clustering of high values (hotspots) and significant clustering of low values (cold spots) of Internet use. The higher the Z score is, the stronger the intensity of the clustering. A Z score near zero indicates no apparent clustering, a positive Z score indicates clustering of high values, and a negative Z score indicates clustering of low values. The red color indicates hotspot areas with Z scores > 0 (3.381411–13.688088), and the blue color indicates cold spots with Z scores < 0 (-3.092929- -1.651054). Based on the Getis-Ord Gi* statistical analysis, significant hotspots of Internet use were found in Addis Ababa, Dire Dawa, Harari, the Northeast part of SNNP (Hawassa), and a few places in Oromia (East Shewa, West Shewa, and East Hararge), Somali, and Amhara (North Shewa). However, statistically significant cold spots of Internet use were found in most parts of the SNNP and some parts of the Benishangul Gumuz, Gambella, Somali, Afar, and Tigray regions of Ethiopia (Fig. 4).

Fig. 4

Hotspot analysis of internet use among reproductive-age women in Ethiopia, EDHS 2016

Spatial interpolations

The ordinary kriging interpolation method was used to calculate the distance from the known point to estimate unknown points/areas. It indicated the point in the ranges of the event occurrences to determine the proportion of Internet usage to unsampled areas. Based on the analysis of ordinary kriging, the predicted Internet usage increases from blue to red-colored areas. The red color shows high-proportion areas of predicted Internet use, and the blue color indicates low-proportion areas of predicted Internet use. Addis Ababa (38.82%), Dire Dawa, and the surrounding area (31.5%) were predicted to be the most prevalent areas for Internet use among women compared to other regions (Fig. 5).

Fig. 5

Ordinary kriging interpolation of Internet use among reproductive-age women in Ethiopia, EDHS 2016

SatScan statistics

Purely spatial analysis scanning for clusters with high rates using the Bernoulli model enabled us to identify specific local clusters, which are necessary for specific local interventions. One primary and eight secondary (a total of 9) were significant (p < 0.05), and Internet use clusters were identified. In total, 57 locations were found in the primary cluster. This spatial window was centered at 9.025638 N, 38.717479 E with a 19.85 km radius, with a relative likelihood of 10.24 and LLR of 425.16, at p value < 0.001. The cluster was located in the Addis Ababa city administration. This means that those who lived in these areas were 10.24 times more exposed to Internet use than those outside of these clusters. The remaining secondary clusters found in northeast SNNP, Oromia, Tigray, Amhara, and Dire Dawa are presented in Table 5; Fig. 6. However, secondary clusters with LLR < 14.329377 were nonsignificant clusters (Fig. 6).

Table 5 Most likely clusters of purely spatial analysis of internet use among reproductive-age women in Ethiopia, 2016

Fig. 6

Purely spatial analysis of Internet use among reproductive-age women in Ethiopia, EDHS 2016

Discussion

The study tried to assess Internet use, spatial variation, and its associated factors among reproductive-age group women in Ethiopia. It was based on secondary data analysis from EDHS 2016, which was a nationally representative cross-sectional study.

The findings of this study indicated that in Ethiopia, the magnitude of Internet use to look up health-related information among reproductive-age women was 4.97%. Women’s age, women’s education, women’s marital status, wealth index, own mobile phone, exposure to mass media, family planning, residence, and regions were significant factors in Internet use. The study identified that the spatial variation in Internet use was significantly varied across the country.

The magnitude of Internet use to look up health-related information among reproductive-age women was 4.97%. This result implies that women’s Internet usage to access health information is low. Relative to the current rapid penetration and use of information technology in the country, this result is much lower. This finding is also lower than different studies conducted in Africa and developing countries. Internet use to access health information among reproductive-age women in the Central African Republic (7.1%), Eritrea (8%), Comoros (8.5%) [15], Poland (66.7%) [13], China (88.7%) [11, 12], and Sweden (84%) [10].

The possible reason for the difference might be due to the variation in study periods, as well as because of the difference in socioeconomic development and the availability of technology between these countries [16]. Additionally, it might be due to access to infrastructure related to technology and national policy related to Internet development programs contributing to the disparity by enabling women to utilize the Internet to obtain health information in a developing and developed country. This might be related to differences in infrastructure, awareness level, and education level between these countries.

The study identified numerous factors that have a significant influence on Internet use for health purposes. The individual- and community-level factors associated with Internet use were predicted using multilevel mixed-effect regression. The findings indicated that women’s age, highest education level, marital status, household wealth index, own mobile phone, exposure to mass media, family planning, residence, and region were significant factors for Internet use.

This finding showed that the age of women has a significant influence on the use of the Internet for health purposes. Women aged 35–49 years were 57% less likely [AOR = 0.43; 95% CI (0.31, 0.61)] to use the Internet for health than women aged 15–24 years. This was supported by previous studies in Saudi Arabia and the United Kingdom [3, 20]. The reason might be that younger women have given more importance to extrinsic rewards of technology use. They have a strong intention to use technology and perceive the usefulness of technology use. Moreover, young women have no large family size, and this could decrease the burden of women, such as workload and economic burden, to meet basic needs, which can increase the use of the Internet to gain information about their health [51]. Similarly, it was supported by studies conducted in Japan and the USA. There is a significant association between age and Internet use for health. Women aged between 15 and 24 years use the Internet more frequently than those aged 40 to 49, and women’s Internet use declines with increasing age [19, 21].

The findings of this study also showed that women’s education has a positive significant association with Internet use. Women with secondary and above education were 6.47 times [AOR = 6.47; 95% CI (5.04, 8.31)] more likely to use the Internet for health purposes than those with primary education. This finding is consistent with studies reported in Columbia [25], the British [24], the University of Arizona-Mexico [23], and Iran [22]. This might be due to educated women being more willing to search the Internet for up-to-date, high-quality health information and having more self-assurance and capabilities to take action regarding their health and the health of their family. Additionally, it might be that women with a secondary and above education have better access to information regarding women’s health than women with a primary education. This could also be educated women who can search and use the Internet, read content, understand the context, and know the service topic [52].

The marital status of women has a significant influence on the use of the Internet for health information. Unmarried women were 2.60 times [AOR = 2.60; 95% CI (1.89, 3.56)] more likely to use the Internet for health than married women. It is supported by evidence in Israel [28]. The possible reason might be that married women spend more of their time with their families than single women. This might also be due to the higher educational status (secondary and above) of single women accounting for a higher percentage (50.83%) in this study. In contrast, the findings of this study in the United States of America [26, 27], in which married women were more likely to use the Internet for health. The adult women who used the Internet to search for health information reported that the Internet had a positive impact on their marriage or partnership. This may be due to married women’s need for support and access to more information about their health as well as the health of their children [53].

Different studies suggest that the wealth status of respondents has a significant influence on the use of the Internet for health purposes. This study also identified that the household wealth index has a significant influence on the use of the Internet for health among reproductive-age women. The likelihood of Internet use among women from rich and middle-sized households was higher than that among women from poor households. Women from rich households were 1.95 times [AOR = 1.95; 95% CI (1.00, 3.80)], and women from middle-income households were 2.39 times [AOR = 2.39; 95% CI (1.20, 4.77)] more likely to use the Internet than women from poor households. This is supported by studies reported in Iran [22], France [30], and China [29]. The possible reason might be that middle and rich households’ wealth index is more likely to have higher incomes, and access to the Internet and having Internet at home can all increase reproductive-age women’s chances of using the Internet for health purposes more regularly. Higher wealth enables people to purchase new technological devices (Wi-Fi, mobile phones) that are financially unreachable to others [54]. In other words, women in poor households are less likely to use the Internet for health because of their economic constraints on meeting their basic needs, while wealthy women are more independent in decision-making, which is supported by studies in China [29].

According to this finding, mobile ownership was positively associated with Internet use for health purposes. Women who had a mobile phone were 3.74 times [AOR = 3.74; 95% CI (2.75, 5.09)] more likely to use the Internet than those who did not have mobile phones. This finding was supported by evidence from Columbia [55] and Japan [19]. This might be due to currently different technologies widely available for use that would ease access and use of the Internet. Having a mobile phone is a portable way of accessing Wi-Fi Internet wherever available at a low cost [31, 56].

Mass media exposure was a significant determinant of Internet use for health information. Women exposed to mass media were 2.63 times [AOR = 2.63; 95% CI (2.03, 3.42)] more likely to use the Internet than women not exposed to media. This is consistent with the findings of other studies in the country [35, 57]. This might be because the media is a beneficial and widely available source of information for disseminating maternal-related health information. The literature has also documented that mass media exposure is a significant source of health information and promotes health-related behavior in women [57]. They might also be encouraged to broaden their knowledge on a health topic further using the Internet once they are aware of it through the media.

In different studies, women-related behaviors affect Internet usage for health purposes. In this study, the odds of women who ever used family planning were 1.38 times [AOR = 1.38; 95% CI (1.04, 1.85)] more likely to use the Internet than women who did not use family planning. This is supported by studies in Indonesia [33] and Europe [32]. The possible reason might be that women who use family planning methods want to choose the best family planning method and to know the side effects of different family planning methods. Additionally, it might be to improve women’s attitudes toward contraception by using the Internet as a common source of information for accessing health-related information [32].

The odds of women living in urban areas were 1.80 times [AOR = 1.80; 95% CI (1.08, 3.01)] more likely to use the Internet than women in rural residences. It is supported by in Sub-Saharan Africa [43], China [36], and the USA [30]. The possible reason could be that the spread of Internet infrastructures in urban settings makes women in urban areas more likely to use it than women in rural areas. In addition, the availability of electric power and women’s exposure to mass media might be one reason that urban residents used the Internet more than rural residents.

The contextual analysis indicated that both city administrations had shown a significant association with Internet use for health purposes. Respondents who lived in Addis Ababa 2.60 times [AOR = 2.60; 95% CI (1.56, 4.35)] and Dire Dawa 2.99 times [AOR = 2.99; 95% CI (1.27, 6.96] more likely to use the Internet than those in the Tigray region. This finding was supported by studies performed in Sub-Saharan Africa [30] and the USA [36]. This might be because city administrations have good infrastructure that provides broadband service, more educated respondents, and a higher economic community than other regions. This might be because Internet service is available in offices, cafes, schools, and public places in the city.

The spatial variation of Internet use for health purposes was significantly varied across the country. The variation in Internet use could guide interventions, as it showed clustering in some parts of the country. The global Moran’s I result was positive and leaned to the right of the curve (skewed to the right). The significant positive Moran’s I was 0.58 with a p value < 0.001 and z score > 1.65, indicating that there was statistically significant clustering (nonrandom) of Internet users in the country. This means there is less than 1% likelihood that the clustered pattern could be the result of random chance.

The Getis Ord Gi* statistical analysis identified the significant spatial clustering of high values (hotspots) and significant clustering of low values (cold spots) of Internet use. The higher the Z score is, the stronger the intensity of the clustering. Based on the Getis-Ord Gi* statistical analysis, a significant hotspot (Z score > 0) of Internet use was found in Addis Ababa, Dire Dawa, Harari, the northeastern part of the SNNP, and a few places in the Oromia, Somali, and Amhara regions. However, statistically significant cold spots (Z score < 0) of Internet use were found in most parts of the SNNP and some parts of the Benishangul Gumuz, Gambella, Somali, Afar, and Tigray regions of Ethiopia.

Purely spatial scan statistical analysis identified the encircled significant areas as Addis Ababa city. This finding is supported by a study conducted in the country [7] and European Union [58]. This might be due to the sociocultural and socioeconomic differences between women in different regions and the difference in infrastructure availability. In addition, the geographic variation in Internet use might be attributable to the difference in awareness and attitude of women toward the positive consequences of acquiring health and health-related information through the Internet [59].

Furthermore, these variations are possibly explained by the wide regional difference in educational status and media exposure. Moreover, in this study, 30.55% of women in Addis Ababa, followed by 12.40% in Dire Dawa city, were exposed to mass media, which is higher than other regional states of Ethiopia. This high educational status and media exposure of the city administration might lead to higher Internet use. This is because high education levels and high media exposure are usually associated with high use of the Internet [59].

Strengths and limitations of the study

This study was based on the EDHS, which is a nationally representative survey with a large sample size that is weighted and can be generalizable to reproductive-age women in Ethiopia. Additionally, the application of advanced methodologies such as multilevel mixed effect regression and the use of spatial analyses helped to detect specific statistically significant Internet use hotspots and cold-spot areas, which was another key strength of the study. This study recognized some limitations, and the interpretation is based on the following limitations. The EDHS 2016 dataset may be taken as outdated. Due to the cross-sectional nature of the survey, a cause-and-effect relationship between outcome and predictor variables cannot be established. The SaTScan analysis detects only circular clusters, and irregularly shaped clusters were not detected.

Conclusion

According to this study, the overall prevalence of Internet use for health purposes among women in the reproductive age group is 4.97%. There was strong evidence of spatial variation in Internet use across the country among reproductive-age women in Ethiopia. Younger women’s age, secondary and above education, single/unmarried, rich and middle wealth status, own mobile phone, exposure to mass media, use of family planning, and living in urban and city administrative regions had a significant influence on Internet use for health purposes among women.

Therefore, improving women’s education, improving access to media, encouraging them to use family planning, and empowering household wealth could improve women’s Internet usage to obtain appropriate health information.