1 Introduction

The household composition has a significant influence on the lives of individuals [1]. It is often determined by as well as determines socioeconomic status, cultural system, religious affiliation, biological and psychological consequences, knowledge of family planning, government policies, and political factors [2,3,4]. Therefore, it becomes essential to identify what determines the size of a household as it affects different sociocultural, political, economic, demographic, and health issues, particularly in a developing country such as Bangladesh.

Bangladesh is one of the most densely populated countries in the world. However, Bangladesh has experienced a decreasing trend in total fertility rate (TFR), and the average size of households has decreased from 4.90 (2001) to 4.06 in 2016 [5]. The household size in Bangladesh is highly influenced by religious beliefs, availability of contraceptives, son preference, extended family, grandparents’ pressure, the continuation of the next generation, security in old age, physical protection, family labor, rural power structure, awareness, and so on. On the other hand, different factors including empowerment of women, family planning program, awareness development, economic hardship, and late marriage have also contributed to the smaller household size. In addition, the government’s slogan, either boy or girl, and two children reflect the institutional endeavor to limit family size in Bangladesh.

Theories about fertility behavior emphasize how structural changes in society lead to changes in birth rates and highlight changes in family size preferences [6]. The neoclassical economic model of fertility behavior, suggested by Becker (1960) and Lewis (1973), had developed a framework in which the demand for children could be treated as durable goods (see [7]). The determinants of fertility work through the demand for children, the potential output of children, and the costs of fertility regulation [8]. Scholars have noted the impacts of income [9], occupation [10], religious affiliations [11], ethnic group [12], parental influence [13], and sex preference [14] on rising family size. Parents consider the costs of having children against the expected economic, social, and psychological outcomes. Wealthy and higher educated people desire to have optimum fertility, while poor, uneducated people have virtually uncontrolled fertility. Parents of rural families desire to have more children as they could provide labor in the rural agricultural sector and also serve as insurance in old age [15].

Women’s reproductive behavior in a given community can be affected by age at entry to marriage, access to family planning, economic status, and cultural and traditional norms in which the woman lives [16]. In 1945, Notestein [17] observed how religious doctrine encourages high fertility, and the power of values and customs could limit the influence of economic development on fertility. Regardless of the indicator used, studies have shown that the desired family size is negatively associated with economic status; couples with a low position desire to have more children than those with a higher position [18]. A substantial amount of empirical literature demonstrates that there is a strong negative correlation between educational level and fertility preferences and behavior.

The existing literature suggests a declining trend in the Bangladeshi population, and a number of influencing factors play a significant role. The number of earning persons, housing type, number of rooms, and persons living in the leading dwelling houses influence family size determination [19]. Parents expect large families because children are highly valued for their contribution to food production and household activities [20]. Similarly, residence, religious affiliation, wealth index, age, education, experiencing child death, and empowerment indicators are significantly associated with mothers’ desired number of children [21]. The discontinuation of contraceptive methods may lead to unplanned pregnancies and unwanted births in Bangladesh [22]. However, to the best of our knowledge, there have been no studies on what determines household size in Bangladesh in recent times, although such research is very important. Therefore, our study aimed to identify the distribution and determinants of household size using the latest nationally representative Bangladesh Demographic and Health Survey (BDHS) data. The findings of this study will benefit researchers in formulating and improvising strategic policy interventions to meet the sustainable development goals (SDGs).

2 Materials and Methods

2.1 Data source

The analysis of this study was performed using a nationally representative survey, the 2017–2018 BDHS data. National Institute of Population Research and Training of the Ministry of Health and Family Welfare managed the dataset, and the survey was carried out by a Bangladeshi Research Institute named Mitra and Associates. The United States Agency provided necessary financial support for conducting this survey for International Development. We used an enumeration area with a mean of about 120 households as a primary sampling unit [23].

A two-stage stratified sampling procedure was applied during the survey period to select the respondents. We chose this particular apparatus because samples were stratified by geographical region and by urban or rural areas within each region. Initially, the smallest administrative units were labeled as enumeration areas (EAs) or clusters and selected with probability proportional to their size. Later, households were selected from each EA. We used a weighted sample of 18,891 individual women who had given birth to at least one living child in the 5-year period prior to the study.

2.2 Outcome Variable

The dependent variable for this study was household size. We categorized this variable into three types i.e., small, medium, and large.

2.3 Explanatory Determinants

In this study, we considered the following determinants as the independent determinants: sex of household head (male, female), division (Barisal, Chittagong, Dhaka, Khulna, Mymensingh, Rajshahi, Rangpur, Sylhet), a decision about health (respondent and husband, others), respondent’s age (≤ 25 or > 25 years), residence (rural, urban), respondent’s education (higher education, secondary, primary, no education), husband’s education (higher education, secondary, primary, no education), wealth index (rich, middle, poor), media access (yes, no), religion (others, Islam), husband’s occupation (businessman and jobholder, agricultural, labor and others, unemployed) and respondent’s working status (yes, no).

2.4 Statistical Analysis

We calculated descriptive statistics of selected determinants and identified the frequencies first. We employed the chi-squared test for bivariate analysis to determine the significantly associated factors with the exposure variable. The average annual rate of increase (AARI) [24, 25] was utilized to observe the household size increasing rate per year. We further applied multinomial logistic regression for this research. This model is used for identifying the relationship between independent determinants and a categorical dependent variable. It also calculates the probability of occurrence of an event by a logistic curve of fitted data. As we categorized our outcome variable into three types, multinomial logistic regression was selected to determine the factors controlling the outcome.

Logistic regression produces the coefficients, including its standard errors and level of significance. The coefficients of a formula for predicting a logistic transformation of the probability of the existence of the characteristic of interest:

$${\text{log}}it\left( p \right) = a_{0} + a_{1} X_{1} + a_{2} X_{2} + a_{3} X_{3} + \cdots + a_{j} X_{j} ,$$

where p is the probability of the existence of the attribute of interest, the logistic transformation can also be defined as the logged odds:

$${\text{Odds}} = \frac{p}{1 - p} = \frac{{\text{Probability of existence of attribute}}}{{\text{Probability of absence of attribute}}}.$$

Alternatively,

$${\text{log}}it\left( p \right) = \log \left( {\frac{p}{1 - p}} \right).$$

The prime distinction of logistic regression is in estimation, which maximizes the likelihood of observing the sample values [26, 27] whereas ordinary regression minimizes the sum of squared errors.

3 Results

Table 1 shows the frequency and percentages of different selected determinants regarding household size in Bangladesh.

Table 1 Baseline characteristics of different selected determinants
Full size table

The results of the chi-squared test in Table 2 showed that division, sex of household head, the decision about health, residence, respondent’s education, husband’s education, wealth index, husband’s occupation, respondent’s working status, and respondent’s age are associated with the exposure variable household size (P < 0.05).

Table 2 Bivariate analysis of household size by different selected determinants
Full size table

Table 3 shows the AARI in the household size in Bangladesh from 2007 to 2017. For the large household size, the AARI in the Barisal division was lowest (− 10.4%), indicating a decrease in household sizes. Male-headed households reported the lowest AARI for larger households (− 3.1%). The AARI for large household sizes was highest if decisions about health were made by the respondent or her husband (0.8%). Similarly, the AARI of urban residence people reported the lowest value for larger households (− 7.7%). Higher educated respondents reported higher AARI for large households, while husbands with no education reported the lowest AARI (− 8.6%) for the bigger households. Also, the AARI was lowest for rich people (− 4.7%), representing a large household size decrease. Households with no media access had the lowest AARI (− 3.1%), meaning a decrease in the size of large households. Unemployed husbands reported the lowest AARI (− 6.7%), representing a decrease in large household size, and other religions had the lowest AARI (− 4.0%) than respondents whose religion is Islam. The AARI for larger household sizes reported the highest value (2.9%) for working women, while respondents over 25 years had the lowest AARI (− 3.9%), representing a decrease in household size. On the other hand, respondents from the Dhaka division showed a higher AARI for medium household size (2.1%). Male- and female-headed households had the same AARI (0.7%) for medium household size in 2007 to 2017. The AARI for medium household size was 3.4% if the respondent or her husband made the decision. Respondents living in rural areas had a higher AARI (2.1%) for medium household size and higher educated respondents, and primary educated husbands had higher AARI, which was 5.0% and 2.9%, respectively, for medium households from 2007 to 2017. Respondents whose wealth index was categorized as poor had an AARI of 1.5% for medium households, and respondents who had media access had a 0.9% AARI for medium households. Muslim respondents had a higher AARI (0.8%) for medium household size compared to non-Muslims. The AARI was higher when the husband’s occupational background was agriculture, labor, and others; that is 1.1% for medium households. The average annual rate for medium household size reported the highest value (5.3%) for working women, while respondents aged older than 25 years had the highest AARI for medium households (1.4%) (see Table 4).

Table 3 Average annual rate of increase (AARI) in the household size in Bangladesh from 2007 to 2017
Full size table
Table 4 Parameter estimation of multinomial logistic regression model for factors affecting household size in Bangladesh
Full size table

Multinomial logistic regression analysis showed that respondents with a male household head had a 2.730 (2.463, 3.026) times higher likelihood of a medium household size than the smaller ones. Similarly, respondents with a male household head had a 4.003 (3.478, 4.606) times higher chance of a large household size compared to the smaller ones. The probability of medium household size compared to smaller ones was smaller for all divisions compared to Sylhet. However, the P-value for Barisal and Chittagong division was not significant for medium household sizes. The probability of medium household size compared to the smaller ones was 0.776 (0.695, 0.844) times lower when the decision about health care services was taken by either the respondent or her husband instead of others in the household. Similarly, the probability of larger household size compared to smaller ones was 0.567 (0.508, 0.632) times lower when the decision about health was taken by either the respondent or her husband rather than the others.

There was a 0.567 (0.519, 0.620) times lower likelihood of medium household size compared to the smaller ones when the respondent’s age was ≤ 25 years. However, the possibility was 1.151 (1.039, 1.275) times higher in large household sizes compared to the smaller ones for the same determinant. The probability of a medium household size compared to the smaller ones in the rural area was 1.332 (1.212, 1.464) times higher than the households in an urban area (P < 0.01). When compared between small households and the large ones, we found that the probability of large household sizes compared to smaller ones in the rural area was 1.919 (1.712, 2.153) times higher than the households in an urban area (P < 0.01). When the respondent’s husband was either a businessman or a job holder, there was 1.505 (1.161, 1.953) times more likelihood of medium household size compared to the smaller ones, while there was 1.356 (1.000, 1.838) times more chance of larger households compared to the smaller ones for the same determinant. The rich people were 0.782 (0.698, 0.877) times less likely to have medium household sizes compared to smaller ones. However, the rich and middle classes were 1.328 (1.159, 1.522) times and 1.234 (1.078, 1.413) times more likely to have larger households, respectively, compared to the poor. The probability of a large household size was 0.796 (0.721, 0.879) times when the respondent was working.

The overall effectiveness of the model was assessed using the chi-squared test. The chi-squared value of 2061.343 and its respective P-value of 0.000 was less than 0.05 (see Table 5). This indicated a significant relationship between the dependent variable and the set of independent determinants in the final model.

Table 5 Multinomial model fitting information
Full size table

4 Discussion

This study aimed to identify the determinants of household size based on data extracted from the BDHS 2017–2018. AARI for household size revealed that the larger household size has decreased during the last decade, but the TFR is still not up to the expected level [23]. The findings of our multinomial logistic regression showed that sex of the household head, division, decision about health, respondent’s age, residence, respondent’s education, husband’s education, wealth index, religion, husband’s occupation, and respondent’s working status were the significant determinants affecting household size in Bangladesh. We found that male-headed households tended to be larger compared to female-headed ones. This finding was consistent with a couple of previous studies from Ethiopia and Egypt [28, 29]. This might be due to the fact that men are often less interested in family planning and care less about the family size, which helps the household size grow bigger. These findings call for an emphasis on propagating awareness programs about family planning methods, particularly among males, to keep the household size smaller in Bangladesh. Dibaba and Mitike [28] also pointed out that women prefer smaller family sizes when maternal education and knowledge of family planning is higher.

Our results indicate that geographic factors influenced the household size considerably in Bangladesh. We found that all of the divisions had a smaller household size compared to Sylhet. This might be due to the fact that family planning programs have been less demonstrated in Sylhet compared to other divisions in the past [30]. Besides, the per-capita income in Sylhet is higher than that in many other areas of the country, and the finding of our study showed that larger household size is common among the wealthy class. This finding raises concerns for an area-specific family planning program focusing on the Sylhet division to reduce the population in this region. There may be additional determinants influencing the distribution of larger households in certain divisions compared to the others, and further research is required to understand those determinants as it is beyond the objective of the present study. Moreover, our study also suggests that household size was significantly influenced by who makes the decision when it comes to health issues. When respondents and their husbands made decisions regarding health themselves, the households tended to be smaller. This indicates the advantages of greater self-control about health-related choices to control the family size positively.

We found the respondent’s age to be a significant predictor of household size, which corresponds with previous studies [19, 31]. Respondents in the younger age group were more prone to having large households compared to their counterparts. The plausible explanation for this finding lies in the biological fact that age is associated with fertility in reproduction function, particularly in women [32]. Besides, the age composition of husband and wife influences the maturity of choice to form a family size. Notably, the influence of age is more prevalent among the wives [33].

Household size further varied based on residence in our study. This finding is in line with a study that indicated a similar proposition [34]. As we observed that household size tends to be larger in rural areas, the plausible explanation for this might be the reason that people living in the countryside lack education and have limited access to different types of services such as mass media [30]. Consequently, rural people have little knowledge about contraceptive use, resulting in increased household size [19]. This finding calls for greater attention to elevate awareness in rural areas and ensure the availability of access to mass media for the promotion of family planning programs. However, Ali [35] did not find any variation in household size based on their location. This may be because of the obsolescence of the study, and there have been tremendous changes in the society and culture of Bangladesh since the survey was conducted.

The results of the present study revealed that respondents with higher education were more likely to have a large household size, which is inconsistent with Pandey et al. [36], who noted that a woman’s fertility is negatively related to the level of education. One possible reason might be due to the fact that higher education for women, in general, is associated with a higher probability of involvement in income-generating activities. As a result, to look after their offspring, they need the assistance of their parents or parents-in-law, which positively influences the household size. Similarly, we found that husbands with primary and higher education were more prone to have larger household sizes compared to husbands with no education. However, this finding was inconsistent with a recent study by Laksono & Wulandari [31], who noted that husbands who have primary education have a higher possibility than husbands with no education to have a family size ≤ 4. This indicates the fact that education influences the understanding of the couple in planning the future of the family. Furthermore, children born in smaller families can often obtain better education [37, 38]. The availability of family-owned resources allocated to a smaller number of family members can be linked to this circumstance [39, 40].

Religion was significantly associated with household size in our study, which matches with some existing literature [41, 42]. We found that non-Muslim respondents were less likely to have larger households compared to their Muslim counterparts. This might be because of the ever-present debate over using family planning methods among Muslims in Bangladesh. Zito and Constantine [33] also noted that intense religiosity is associated with larger family sizes. We found a significant role of wealth in determining household size, which corresponds with Begum [19]. Respondents from the rich and middle class had a higher probability of larger household sizes in this study. This might be due to the perception that they have enough assets to support more children, consequently continuing to expand the family size.

A husband’s occupation was found to be a significant determinant of household size in our study, which corresponds with Pandey et al. [36]. The possibility of a larger household size was higher when respondent’s husband was either a businessman or a job holder than others. However, Adamchak and Mbizvo [43] did not find any such link in their study. We also noted that working respondents were more likely to have smaller households. This is because working women tend to be more aware of the limited time available to care for children, so they prefer a smaller family. However, this finding contradicts the study of Adamchak and Mbizvo [43], who reported that unemployed women had significantly fewer children than women working in agriculture. Laksono and Wulandari [31] and Pandey et al. [36] reported a similar finding from their study in Indonesia and India, respectively [31].

4.1 Strengths and Limitations

The main strength of this study is that we used the nationally representative data extracted from the latest Bangladesh Demographic and Health Survey 2017–2018. The large number of random data used for this research makes the study more generalized, valid, and reliable. We examined the determinants of household size from the dimensions that are less mentioned in previous studies. Besides, we used random sampling methods and multinomial logistic regression to increase the validity and reliability of our results. The findings are largely consistent with most of the previously existing literature, with a few contrasts. We suggested some important policy implications based on our findings from the recent BDHS dataset representing different society segments.

While the present study has specific strengths, we acknowledge the limitations of our study. We focused on a set of specific determinants determining household size for our analysis. This study did not include age at marriage [35, 44], sex preference, child mortality [34], entertainment, family culture background, biological factors [44], technological factors [45], mothers’ health and psychological factors [28], contraceptives using [46], and income [35], as mentioned in other studies. Additionally, the present study is quantitative in nature, and the data were collected from a secondary source, making it difficult to triangulate.

5 Conclusion

The study of what determines household size is essential in the context of Bangladesh. After analyzing the recent nationally representative BDHS data, we found that sex of the household head, division, decision about health, respondent’s age, residence, respondent’s education, husband’s education, wealth index, religion, husband’s occupation, and respondent’s working status were the important determinants affecting household size in Bangladesh. However, to our surprise, we did not find any significant link between access to media and household size in our study. Based on our analysis, we recommend that involving women in decision-making processes such as family planning, especially in rural areas, can be helpful controlling the family size. Besides, promotion of education for women as well as female involvement in income-generating activities should be encouraged to create awareness and women empowerment, which can help control the population. We suggest further qualitative research for triangulation and quantitative analysis, including additional determinants to understand this topic comprehensively.