2. Literature review

At the end of the 20th century, several low-income countries still featured high population growth rates. Ever since, fertility rates have declined almost everywhere. Between 1996 and 2016, the fertility rate in South Asia declined from 3.8 to 2.5 children per woman. In Latin America and the Caribbean, it went from 2.8 to 2.1, in sub-Saharan Africa the decline has been from 6 to 4.8 children.

Understanding the causes and channels that trigger the fertility decline is a crucial element for introducing effective policies. In fact, family planning programs were often thought to be less successful in reducing fertility compared to their originally stated target [Caldwell and Caldwell (Reference Caldwell and Caldwell2002)]. In Pakistan, family planning programs were first introduced by Begum Saida Waheed, in 1951, who founded the Family Planning Association of Pakistan (FPAP) when her maid died during pregnancy. During its second five-year-plan (1960–1965), 1,589 family planning clinics with healthcare dispensaries, hospitals and maternal and child centers were set up. “Lack of motivation in the general masses, unsatisfactory arrangements for the distribution of contraceptives, inadequate education and information of the masses in family planning […]” were identified as the main causes behind a low impact of the family planning program in Pakistan in those years [Ali and Zahid (Reference Ali and Zahid1998)]. Only after a series of eight family plans, a fertility decline was finally recorded, and it coincided with changes in the socio-cultural role women had within the family. Pakistan's “Population Policy”, introduced in 2002, proved to be not successful either, mainly due to the fact that it rested on the involvement and support of Imams but struggled with muted opposition [Brohi and Zaman (Reference Brohi and Zaman2017)]. On the other side, the Lady Health Workers Programme (LHWP) did contribute to a reduction in the fertility rate in Pakistan [Sultan et al. (Reference Sultan, Cleland and Ali2002), Khan and Wang (Reference Khan and Wang2021)]; launched in 1994 by Pakistan's Ministry of Health, it provided doorstep family planning services through literate women who had undergone a 15-month training. By 1996, approximately 30,000 women were working under this program and 110,000 women in 2013 [Khan and Wang (Reference Khan and Wang2021)].

A scholarly consensus on the primary drivers of fertility decline does not exist. Pritchett (Reference Pritchett1994) argued that the family planning centers started off the fertility decline, while other scholars praised the roles played by family planning messages presented on television [Bongaarts et al. (Reference Bongaarts, Mauldin and Phillips1990), Casterline and Sinding (Reference Casterline and Sinding2000), Bongaarts (Reference Bongaarts2006), Bongaarts and Casterline (Reference Bongaarts and Casterline2013)]. The electricity-fertility nexus is contested, too. From the neoclassical microeconomic tenets, electricity may substitute for child labor, thus increasing the net costs of raising children. If electricity increases the demand for skilled labor the related increase in the returns to education may incite parents to substitute away from a large quantity of children and spend more on their “quality” [Dinkelman (Reference Dinkelman2011), Barron and Torero (Reference Barron and Torero2014)]. Supply-side theories of fertility, in turn, predict that electricity boosts fertility, as it rises the quality of medical care which may reduce both fetal and child mortality [Easterlin and Crimmins (Reference Easterlin and Crimmins1985)]. However, existing evidence suggests that hoarding is only practiced as long as parents anticipate that some of their children will not survive until adulthood and overcompensate with high fertility [Sah (Reference Sah1991), Schultz (Reference Schultz, Rosenzweig and Stark1997)]. The early studies on the electricity-fertility nexus –published in the 1980s– find a negative relationship [Herrin (Reference Herrin1979), Harbison and Robinson (Reference Harbison and Robinson1985), Cornwall and Robinson (Reference Cornwall and Robinson1988)]. More recent studies such as Potter et al. (Reference Potter, Schmertmann and Cavenaghi2002) and Bailey and Collins (Reference Bailey and Collins2011) confirm the negative effects electrification had on fertility for the case of Brazil and the USA. Burlando (Reference Burlando2013) exploits the effect of a months-long power outage across electrified and non-electrified villages in Zanzibar and finds the lack of electricity to be positively related with fertility rates. Peters and Vance (Reference Peters and Vance2011) find contrasting effects for urban and rural areas of Côte d'Ivoire: electrification seems to decrease fertility for urban households.

The role of the mass media in changing fertility behaviors has received attention more recently. Television not only affects fertility choices but also influences people's consumption patterns, religious beliefs and socio-political views [Johnson (Reference Johnson2001)]. Jensen and Oster (Reference Jensen and Oster2009) show that after the introduction of cable television in rural India, the respect toward women by other family members increased. Recent studies on the role of television in developing countries such as Jensen and Oster (Reference Jensen and Oster2009) focus on the content presented on television. La Ferrara et al. (Reference La Ferrara, Chong and Duryea2012) demonstrate that the expansion of soap-operas in Brazil contributes to the fertility decline and, in a similar vein, Banerjee et al. (Reference Banerjee, La Ferrara and Orozco2017) demonstrate that Nigerians exposed to a tele-serial on HIV/AIDS had an increased likelihood of getting tested for HIV.

3. The industry of television series in Pakistan

Television arrived in Pakistan in 1964 when the state-led Pakistan Television (PTV) was launched. Until the early 1990s, PTV enjoyed a near-universal monopoly over the airwaves. At the turn of the century, privately owned commercial channels were licensed. Consequently, during the 2001–2010 period the television spectrum expanded exponentially. On the one hand, the number of channels increased from 4 in 2003 to 89 by 2010 and, on the other, physical outreach expanded at an unprecedented pace.

In 2018 Pakistan, 98 television channels were fully functional including 38 in the entertainment category (music, sports, and kids' channels). Television has emerged as the most important mass-media tool. Two-thirds of the population watch television, whereas the internet reaches only 20% of the people. One in five reads newspapers while there has been a 39% decrease in radio listening [Gallup (2018a)]. The situation was similar five years ago: 7% listened to the radio, 18% read newspapers but 73% spent time watching television [Orient (2013)]. On average, a Pakistani person spends 115 min per day watching television (Gallup, 2018a); women and those under the age of 17 spend more time watching entertainment channels while men watch news and sport channels [Gallup-Pakistan (2018b)]. The audience of news channels consists of two-thirds of men, entertainment channels (with the exception of ARY Digital) attract more women. For instance, the entertainment channels Hum, Urdu 1 and Geo Entertainment have, respectively, 61%, 56% and 54% of their audience constituted by women [Gallup-Pakistan (2019)]. This reflects the interest of women and girls in soap-operas.

Moreover, the slogan “Kum bachchay khushal gharana” (Fewer children, more prosperity) was promoted via family planning commercials in the mass media in the 1980s. From 1990s onwards, television has also featured contraceptive campaigns by state institutions such as provincial population welfare departments, as well as NGOs and manufacturers of contraceptives. The advertisements for contraceptive products have become more explicit; as an example, one advertisement features two prominent television actresses showing an intrauterine device while another one explicitly talks about menstruation. At the same time, campaigns to promote contraceptives have faced hostility by the Pakistani state. The Pakistan Electronic Media Regulatory Authority (PEMRA) –a federal body that regulates electronic media– banned all advertisements for contraceptive products in 2016 arguing that such campaigns exposed children to the subject of sex [Hashim (Reference Hashim2016)]. However, the ban was lifted 3 days after criticism by the civil society. Now, commercials about contraception are only aired after 11 pm. Similarly, in 2013 PEMRA banned all television advertisements making reference to condoms. Not only advertisements and talk shows would discretely promote birth-control, contraception and the two-children policy, but even television series reinforce the two-child message. In what follows we briefly introduce the content of three famous and much-watched television series: Alpha Bravo Charlie, Azar ki Aye Gee Barat, and Ishq-e-Mamnooh.Footnote ⁴

Aired in 1998 on PTV, action-thriller Alpha Bravo Charlie (ABC) was a huge success in terms of popularity [Ahmad (Reference Ahmad2015)]. ABC's cumulative reach in 1998 was 91% (remained 66% when re-telecasted in 2006). The series revolves around three military officers: Kashif, Fraz and Gulsher. While Kashif and Fraz represent Pakistani elite, Gulsher comes from a humble background. Kashif is the son of a general and has another brother, Rashid, who is studying in the US. His fiancée is the only daughter of another general. Likewise, Fraz is the only child, and his father is a landlord. In the case of Gulsher, the series remains vague. However, a comical figure in the series, Rasaldar (Corporal), has 12 children and his fellow corporal Jaffari has 13 children. In short, the main characters representing Pakistani elite have a nuclear family with two children at most. In contrast, the working-class families have a dozen children.

The next television series under scan, Azar ki Aye Gee Barat (2009), aired on Geo Entertainment, was such a huge success that Geo aired three sequels in subsequent years [Gallup (2018a)]. This series depicts two upper-middle-class families: the Chaudhry family and the family of Ahmeds. Both families have two children and are connected through intermarriages. Most importantly, the life of the women in the serial is not restricted to the idiomatic four walls. Mrs Ahmed is a successful entrepreneur and Mrs Chaudhry, after a failed attempt to establish a business, joins politics and is elected as a member of the National Assembly.Footnote ⁵ Likewise, most young women in the series have a lifestyle that can be described as “westernized” in the Pakistani context.

The third series, Ishq-e-Mamnooh (Forbidden Love), aired on Urdu 1 in 2013, was a Turkish import and biggest ever success in terms of Television Rating Points (TRPs). It was the first-ever Turkish soap-opera aired in Pakistan. Ishq-e-Mamnooh's plot revolves around two core families, the families of Adnan Ziyagil, a widower, and Firdevs Yoreoglu, a widow. Adnan is a rich and successful entrepreneur living in a luxurious Istanbul villa by the shore of Bosporus Strait and he has two children. He is attracted to Firdevs who has two young daughters, Peyker and Bihter. At Peyker's wedding, Adnan falls in love with Bihter and the two ultimately get married.

A key feature of most Pakistani broadcast series is that the family size depicted there is smaller than the average Pakistani family; in addition, women represented in the series are more westernized and outgoing compared to the traditional image of a Muslim woman. Those clashes between real life and television-based life generated a plethora of criticisms as it was argued that Turkish produced series were perceived to be vulgar and obscene for the Pakistan standards [Sulehria (Reference Sulehria2018)].

4. Theoretical framework

Understanding the causes and channels that may trigger the fertility decline is a crucial element for introducing effective policies. Many attempts have been made to theoretically and empirically explain the causes of the decline in fertility in developing countries. While acknowledging such critical concepts as decoloniality [Mignolo and Walsh (Reference Mignolo and Walsh2018)], our aim is not to mechanically apply Western development models on our case study. There is consensus that part of the decrease can be attributed to the rise in household income, improved literacy, the improved status of women, the decline in infant/child mortality, and the erosion of traditional authorities [Becker and Lewis (Reference Becker and Lewis1973)]. Yet, there are examples where fertility declined even if those conditions did not materialize, and others where fertility did not decline despite the presence of these pre-conditions [Bongaarts (Reference Bongaarts2006)]. For example, Spolaore and Wacziarg (Reference Spolaore and Wacziarg2019) studied the fertility decline in Europe between 1830 and 1970, and highlighted the role played by linguistic affinities/distances across European countries. The authors argue that “fertility decline resulted from the gradual diffusion of new fertility behavior from French-speaking regions to the rest of Europe”.Footnote ⁶

An additional social change that has received scientific attention in explaining the fertility decline is electricity. While electrification is associated with improved economic well-being, it also plays a key role in spreading mass media, particularly radio and television. Television, in turn, can be used to amplify family planning messages at a mass level. Thus, the three channels that we assess in this study as contributing factors to the fertility decline are (i) electricity, (ii) television and (iii) family planning commercials (Figure 2). The first channel, i.e. electricity, influences the costs of raising children since it replaces manual and time-consuming household chores with electric appliances. Moreover, electricity creates new business opportunities as well as alternatives to sexual activity, reduces the need for labor support from children and contributes to an amelioration of health care provision resulting in a decrease of child mortality.

Figure 2. Links between electricity and fertility choice.

Notes: Authors' elaboration.

This leads us to the second channel: television. The effect television may have on fertility outcomes derive from two sources. The direct impact is the exposure to the nuclear family and to emancipated women. Exposure to television changes the marginal utility of having children by broadcasting new lifestyles. La Ferrara et al. (Reference La Ferrara, Chong and Duryea2012) show that soap-operas influences women's fertility preferences. Moreover, exposure to soap-operas is also associated with female empowerment. According to the National Institute of Population Studies (2013) women living in a household without a TV tend to be more likely to justify their husbands' violent behavior than women living in a household with a TV. Put differently, the indirect impact of television is not exclusively linked to fertility. It represents a change in habits shaped by television content.

The third channel is represented by the family planning commercials. More information about family planning services helps to reduce unwanted births by addressing the unmet need for contraception and reducing unwanted fertility because of information on use and access to contraception [Bongaarts (Reference Bongaarts1993)]. Thus, taken together we expect these three channels to have a non-negligible impact on fertility choices.

5. Data

The PDHS conducted four times in 1990–91, 2006–07, 2012 and 2017–18, have been deployed for our analysis. Particular importance –in the context of the four surveys– has been given to the collection of data on fertility levels [Tasciotti and Wagner (Reference Tasciotti and Wagner2018)]. The sample design adopted across waves is a stratified, clustered and systematic sample of households. The sampling frame consists of urban and rural areas of the four provinces of Pakistan, as defined in the population census available at the time of the different surveys. Thus, the dataset is representative of the country at regional, urban and rural levels.

We have complete information on 5,418 women in 1990–91, 8,123 in 2006, 7,437 in 2012, and 12,602 in 2017–18. Our sample includes mothers aged 15–49, i.e. all women who responded to the fertility questionnaire have at least one child. Childless women are excluded as data on their fertility choices is not available. Yet, this is not a concern for the study at hand since childlessness is hardly observed in Pakistan, in fact it is seen as a curse [Qamar (Reference Qamar2018)] and only 5% of the women are childless when reaching the end for their reproductive age [PDHS (2013)]. According to the literature, population-level childlessness has been stable over time, and reported to be between 2%, 4% and 5% in the period between 1980 and 2019 [Alam et al. (Reference Alam, Irfan, Farooqi, Sheikh, Siyal, Ahmad, Nasir and Haq1983), PDHS (2013), Ahmed et al. (Reference Ahmed, Khan, Yasmin, Jawaid, Khalid, Shigri, Nawaz and Hasan2020)].

We have individual-level background information about age and education. We complement the information about the woman with partner characteristics. We know the husbands' age, education, occupation status and whether he lives with his wife. Finally, we can classify all individuals in our sample into one out of five wealth categories based on their asset ownership and identify their location of residence as urban or rural.

Table 1 presents the descriptive statistics of these background characteristics. Since the fertility history is conditional on age, we classified the women into age cohorts of 5-year intervals starting from the age of 15. Most respondents (22.9%) are between 25 and 29 years old followed by another fifth of the sample being between 30 and 34 years old. Across survey rounds the age distribution is coherent. On average, women tend to have 3 years of schooling yet the standard deviation is substantial (4.8) and we observe a considerable increase in the average years of schooling from 1.8 in 1990–91 to 4.3 in 2017–18. There is a considerable age difference between women and their male partners. The average age of the husband is almost 47 years, a difference that is constant across survey rounds, and husbands have spent twice as many years at school compared to their wives. Again, there is a positive trend in schooling, resulting in an additional 2 years of schooling on average for men between 1990–91 and 2017–18. Across survey rounds, almost all husbands are working (96%) and the majority (88%) co-reside with their wives, with a slightly decreasing trend over the years. On average 45% of the interviewees live in urban areas.

Table 1. Descriptive statistics of the background characteristics

Note: Authors' calculation based on DHS surveys from 1991, 2006, 2012 and 2018.

As outlined in the theoretical considerations we focus on three pathways for reducing fertility: (i) access to electricity, (ii) ownership of a television, and (iii) exposure to family planning on television. At the bottom of Table 1 and in Figure 3 we show the evolution of these three channels across survey rounds. Undoubtedly, there was a considerable increase in access to electricity from 1990–91 to 2017–18, with 90% of the households now having access to it compared to 73% thirty years ago. Television ownership experienced a sharp increase between 1990–91 and 2006 from 38 to 58%; the share of households owning a television has stabilized around 60% in 2017–18. While access to electricity and television ownership has been on the rise throughout the entire period under study, the share of households who watch family planning commercials was highest in 2006 (39%); in 2018 only 1 out of 5 households reports watching family planning commercials. In the 1990s family planning messages on television were prominent albeit not necessarily very explicit. In later years the advertisements became more explicit, which is reflected by the 2006 data. However, this trend reversed between 2006 and 2018 since the PEMRA became stricter and more critical about family planning campaigns. This reflects a general trend of increased religiosity in the country that disapproves of any form of birth control. Therefore, it is not surprising that in 2018 a smaller share of women indicate that they have ever heard about family planning on television compared to 1991.

Figure 3. The channels of interest: evolution of electricity access, television ownership and having heard of family planning on TV (1990, 2006, 2012 and 2018).

Note: Authors' calculation based on DHS surveys from 1991, 2006, 2012 and 2018.

In terms of outcomes of interest, we focus on realized fertility as contrary to desired fertility since the former is expected to respond most to social pressures. We consider (i) the crude birth rate, and (ii) the total number of children born alive. Other papers on the same topic tend to consider only one outcome; La Ferrara et al. (Reference La Ferrara, Chong and Duryea2012), for instance, assess the average number of live births among women of reproductive age. Jensen and Oster (Reference Jensen and Oster2009) study the probability of pregnancy. Assessing two related fertility indicators allows us to present a more complete picture of the evolution of fertility choices and their determinants over time and to study whether any found impact is mutually reinforcing and coherent across outcomes.

Figure A.1 (in the Appendix) gives an overview of how the crude birth rate and the number of children being alive changed over time and across categories of age. The figure highlights some trends that are worth to be pointed out. The average number of children per woman exhibits a slow but consistent decline, which is in line with national trends [National Institute of Population Studies (2013)]. The difference between the total number of children ever born and the total number of surviving children diminishes over time with the exception of women aged 35 and older in the year 2012. These statistics indicate the availability of better health infrastructure and a decrease in child mortality over time [National Institute of Population Studies, (2013)]. Importantly, these micro-level data are consistent with the macro-economic data presented in Figure 1. After having established these co-moving trends, i.e. of a reduction in fertility and an increase in access to energy and TV (and to a lesser extent in family planning commercials, rather stagnating), we now set up a multivariate model aiming at identifying whether there is a relationship.

5. Empirical model

One of the main methodological problems affecting this type of research is the difficulty in assessing the causal effect between electricity, television and family planning advertisements on the fertility choice. In other words, establishing a causal effect is technically challenging with the data at hand since we rely on a repeated cross-section.

We adopt a two-fold strategy; first, we employ age-cohort, time and location-specific effects to address the structural aspects that feed into the decision-making process about fertility. We complement the multi-level fixed effects analysis with an IV analysis that treats our three channels as being endogenous. As IV we employ the average yearly district-wealth level access to the respective channel. We opted for the construction of a time-varying district-wealth specific instrument to be able to jointly account for all structural effects that we have already in the fixed-effects model. We employ an IV to address the concern that an omitted variable affects both the number of children and the use of new technology, i.e. electricity and TV ownership as well as exposure to family planning commercials. We opted for the average yearly district-wealth level access to the three channels (electricity, TV, family planning commercials) as instrumental variables as they represent peer-group effects. The peer group is constituted by the wealth quintile, as we expect individuals to compare themselves mainly within their wealth group. This is also what we observe in the data; within the same district, the richest households are the first adopters whereas those in the lowest wealth quintile are the latest adopters. The peer group adoption effect -i.e. the aggregate effect- triggers individual adoption and ultimately individual exposure to the new technology. Please note that aggregate adoption of the technology itself is not directly related to individual realized fertility. It is for that reason that we focus on realized fertility and not on fertility perceptions. In the context of realized fertility, we consider the aggregate measure of adoption within a peer group to be a valid instrument. It reflects the stage of diffusion across wealth quintiles. By contrast, it does not directly determine a single couples fertility realizations, i.e. their actual number of children. We argue that aggregate peer-level adoption only triggers individual adoption leading to individual exposure. It is this individual exposure that has the potential to affect long-held, individual perceptions and ensuing fertility realizations. Thus, by purging individual exposure from aggregate peer-level adoption we can causally attribute the effect of electricity, TV and family planning commercials on realized fertility conditional on the other covariates. Importantly, this instrument takes economic conditions into account and allows us to differentiate by economic groups. Since we also include wealth group-specific effects directly as control variables in the main analysis, we capture both, the direct wealth effect and the access to energy-wealth nexus [compare La Ferrara et al. (Reference La Ferrara, Chong and Duryea2012)]. Our choice of instrument was further motivated by Fisman and Svensson (Reference Fisman and Svensson2007) and Jain (Reference Jain2021) who set up a similar IV approach in the context of firm bribery payments and ICT adoption, respectively. A similar line of argument is also made by Bertschek and Niebel (Reference Bertschek and Niebel2016) and Michaels et al. (Reference Michaels, Natraj and Van Reenen2014) who set up their models in the context of firms. Like the other studies, we consider the instrument valid because of the different levels of aggregation, i.e. the average group level adoption versus the individual realization of the outcome.

Since our outcome variables are count variables, we implement a Poisson model. For consistent estimation, we only have to impose that the conditional mean is correctly specified [Wooldridge (Reference Wooldridge2010)]. Thus, we define the conditional mean for every women i, living in household h, in enumeration area e, at time t as follows (equation (1)):

(1)$$\eqalign{& E( {Fert_{ihet}\vert Chan_{het}, \;\;Ind_{ihet}, \;\;Hub_{ihet}, \;\;HH_{he}, \;\;\lambda_e, \;\;t_t} ) \cr & = exp( {\alpha 0 + \alpha 1Chan_{het} + \alpha 2Ind_{ihet} + \alpha 3Hub_{ihet} + \alpha 4HH_{het} + \lambda_e, \;t_t} ) } $$

where exp(.) refers to the exponential function, Fert_ihet denotes one of the two fertility outcomes we study (children ever born, children born alive). The three channels (Chan_het) under study are first separately and then pair-wise tested. In addition, all specifications include the following individual characteristics (Ind_ihet): age categories and years of education. The husband characteristics (Hub_ihet) are age, education, whether he is working and whether he co-resides with his wife. The household-level characteristics (HH_het) consist of the wealth categories and the urban location indicator. In addition, we control for location-specific effects (λ_e) referred to as enumeration areas and for survey-year specific effects (t_t). Note that the location-specific effects are a key component in our set-up since they allow us to indirectly account for the topography of the different places and their geographic location.Footnote ⁷ The remaining error term is clustered at the enumeration area level since we expect the error terms to be correlated within areas as individuals living close to each other are likely to be exposed to the same shocks and the same unobservable conditions.

In the IV estimation, we complement the above set up with the average yearly district-wealth level access to the respective channel as an instrument. On top of the instrument being motivated by the existing literature, it is also motivated by earlier studies about Pakistan that established a relationship between income and access to electricity [Gellerson (Reference Gellerson1982), Majid (Reference Majid2013)]. In turn, we abstain from using a time-fixed IV such as distance to the capital. Due to the topography of the country the electricity provision is de-centralized, some areas are provided with hydro-electric power, and the placement of these does not depend on the distance to the capital but natural conditions.Footnote ⁸ Similarly, the nuclear power plants are distributed all over the country.Footnote ⁹

The challenge with a two-stage Poisson model is that the variance-covariance estimator in the second stage has to account for the estimates from the first stage. To circumvent a two-stage procedure we employ the control-function estimator of the Poisson model. The approach combines the moment conditions from the two stages in a single GMM estimator making adjustments to the second stage variance-covariance estimator redundant [Newey (Reference Newey1984), Wooldridge (Reference Wooldridge2010)]. Moreover, the control function estimator of the Poisson model augments the original model with an estimated term that controls for the endogeneity of the respective fertility channel under study. We refer to it as ρ. The coefficient estimate associated with the control term measures the strength of the endogeneity. When the fertility channel is exogenous, then the coefficient associated with this control term is zero. Importantly, since we have a time-varying IV that also varies within enumeration areas, all other control variables as specified for the basic Poisson model are kept in the two-stage model as well to allow for a comparison of the models based on the covariates.

One downside of applying the Poisson model is that the coefficient estimates are not directly interpretable as marginal effects. We have to calculate the incidence rate ratio, which depicts the relative risk. We obtain the incidence rate ratio by applying the exponential function to the Poisson coefficient.