Data
This study primarily used the fourth round of National Family Health Survey (NFHS) that was conducted in 2015–16. However, to see changes over time in the level of covert use of modern methods of contraception, the third NFHS (2005–06) round was also analysed. The National Family Health Survey (NFHS) is a large-scale, multi-round survey, conducted in a representative sample of households throughout India. In the selected households, all the eligible women aged 15–49 years were interviewed. Eligible men aged 15–54 years were interviewed only from a fraction of households selected for men’s interview. In the fourth round, the mandate of the NFHS, for the first time, was also to provide a majority of health and demographic indicators at the district level along with state and national levels. However, to save the resources, the state module, comprising men’s interview schedule and Sections 8–11 of women’s interview schedule, was canvassed only in 15% of the total households sampled in the fourth round. In contrast, men’s and women’s interview schedules were canvassed in all the sampled households in NFHS-3 (2005–06) since the mandate was to generate only state and national level indicators. Both the rounds of NFHS (2005–06 and 2015–16) were cross-sectional in nature and provided independent samples of couples in the couple data files, which were analysed for this study [35, 36].
Selection of couples
The present analysis estimated covert use of modern reversible contraceptive methods that include pills, intrauterine devices, injectables, and other modern contraceptive methods for females. The analysis excluded couples among whom women either did not report using any method or reported using traditional methods or female/male sterilization. The other categories excluded were couples among whom either the man reported having more than one wife or the woman reported an additional wife of the husband, or where a man’s last sexual partner was not his wife since the questions asked to men on their contraceptive use pertained only to their last sexual partner. The study is based on 4,825 and 7,824 fecund, monogamous couples from NFHS-3 (2005–06) and NFHS-4 (2015–16), respectively, who were not sterilized and were users of modern reversible contraceptive methods. A few cases in which women accepted using a contraceptive after the last coitus were dropped from the study.
Generating sampling weights for couples
In the DHS survey, men and women from the sampled households were interviewed separately. The DHS program provides men and women data files separately, with individual and household sampling weights. A couples’ data file on matched partners is also available but without the sampling weights for the couples. This issue cannot be bypassed by applying women’s or men’s sampling weights as doing so creates a significant bias in the results. The solution is to develop sampling weights for the couples to make the results representative [37]. Accordingly, normalized couple sampling weights were generated for this study based on the methodology suggested by Becker & Kalamar in 2018 [37]. Another study done in the Indian context by applying couples’ level weights but did not use an appropriate method for the weights’ calculation [38].
Dependent variable
Since NFHS-4 (2015–16) did not directly ask a woman whether ‘your husband/partner knows about your contraceptive use’, the estimate for this dependent variable was indirectly obtained as suggested by Gaska & Becker in 2018 [1]. The survey asked women a series of questions like: ‘Are you currently doing something, or using any method, to delay or avoid pregnancy?’ If the answer was ‘Yes’, then a subsequent question was asked: ‘Which method are you using?’ For men, the question was asked more specifically as: ‘The last time you had sex, did you or your partner use any method (other than a condom) to avoid or prevent a pregnancy?’. The term ‘other than a condom’ was used within parentheses since a question regarding the use of condom at the last coitus was asked to men in a previous section of the questionnaire.
The dependent variable, ‘covert contraceptive use’, henceforth may be called ‘CCU’, is defined as the use of a female modern contraceptive method for spacing as reported by a woman but not by her husband. By this definition, if a woman reports using any female contraceptive method and a man reports using condoms, it is considered a case of open contraceptive use. If husband and wife report different female modern methods, that too is regarded as open contraceptive use.
Independent variables
Based on the literature, a list of independent variables regarded to exert a direct or an indirect influence on covert contraceptive use were included in the analysis. The independent variables were categorized into two parts. The first category included women’s background characteristics, women’s education, religion, caste, place of residence, and wealth [5, 33, 39] while the second category included couple-level variables. Religion, that was included in the first category, was divided into three categories: Hindu, Muslim, and Other. Caste of women was also taken as an independent variable and categorized as: Scheduled Caste/Scheduled Tribe (SC/ST), Other backward class (OBC), and Other. Place of residence was divided into rural and urban sub-categories. Women’s economic condition was categorized into household wealth quintiles as: poor, middle, and rich. Household wealth quintile was taken as a proxy for access to resources in making decisions in general and using contraceptive in particular.
In order to control the regional variation, we included region as an independent variable and categorized states and union territories into different regions based on their geographical location. We adopted the standard categorization as per the NFHS. This categorization, to a great extent, also takes into account sociocultural and demographic variation, marriage and reproduction related customs, women’s status and the overall governance in implementing family planning program that may affect CCU.
On the women’s autonomy front, a battery of independent variables was inserted. Among them, women’s education included the following categories: non-literate, primary, secondary, and higher education. Other independent variables included women’s working status, that is, whether a woman had money she could use independently, and women’s freedom of mobility to visit health facility and market places. The literature suggest that these variables affect women’s contraceptive use as well as the covert contraceptive use [21,22,23]. Both variables were dichotomous, with ‘yes’ and ‘no’ responses.
We also included three couple-level variables. The first one was age difference between husband and wife, which was divided into the following categories: wife is the same age as husband or older, husband is 1–3 years older, husband is 4–6 years older, and husband is much older, which included couples where husband was 6 or more years older than wife. The second variable was educational difference between spouses and was divided into three categories, namely ‘both are equally educated’, ‘wife is more educated’, and ‘husband is more educated’. Education and age gap were included with the reasoning that a spouse senior by age or education tends to dominate in decision-making regarding reproduction and family planning [40,41,42]. We included concordance on desire for more children as the third couple-level independent variable and divided it into two categories, namely ‘yes’ (indicating concordance) and ‘no’ (indicating absence of concordance).
Husband’s egalitarian gender attitude was taken as the main explanatory variable in this study. The working mechanism of this variable is unique and explained below in detail.
Measurement of husband’s egalitarian gender attitude
Husband’s egalitarian gender attitude was measured from the responses to questions posed to men on their perception of various situations in NFHS-4. In the first battery of questions, husbands were asked if beating or hitting wife is justified in the following situations: a) if she goes out without telling husband b) if she neglects children c) if she argues with husband d) if she refuses to have sex with husband e) and if she doesn’t cook food properly. In the second battery, husbands were asked whether wife is justified in refusing sex if she knows: a) husband has a sexually transmitted infection (STI) b) if husband has sex with another woman c) and if wife is tired or not in the mood. In the third series of questions, husbands were asked if, in the situation that wife refuses sex, husband has the right to: a) get angry, b) refuse to give her money or other financial support, c) use force to have sex, d) go and have sex with another woman. The responses to each of these questions were coded as ‘0’ if the attitude was negative and ‘1’ if it was positive. This study used the Latent Class Analysis (LCA) to measure husband’s egalitarian gender attitude. LCA is a statistical procedure used to group individuals into classes of an unobserved variable based on responses made on a set of nominal or ordinal observed variables. It can identify subgroups of individuals who share common characteristics in such a way that they have a similar scoring pattern within the group [43]. The mathematical equation for LCA is as follows:
$${\varvec{P}}\left({\varvec{Y}}={\varvec{y}}\right)=\sum_{{\varvec{c}}=1}^{{\varvec{C}}}{{\varvec{\delta}}}_{{\varvec{c}}}\prod_{{\varvec{i}}=1}^{{\varvec{m}}}\prod_{{\varvec{j}}=1}^{{{\varvec{n}}}_{{\varvec{i}}}}{{\varvec{\tau}}}_{{\varvec{i}}{\varvec{j}}{\varvec{c}}}^{{\varvec{I}}({{\varvec{y}}}_{{\varvec{i}}}={\varvec{j}})}$$
where j = 1, 2, … ni represents the number of possible categories the variable yi can take and I(yi = j) is the indicator function that equals to 1 if the response is y = j and, 0 otherwise. \({\varvec{\delta}}\) c is the probability of membership in the latent class c and the sum of probabilities for all classes equal to 1. \({{\varvec{\tau}}}_{{\varvec{i}}{\varvec{j}}{\varvec{c}}}^{{\varvec{I}}({{\varvec{y}}}_{{\varvec{i}}}={\varvec{j}})}\) is the probability of variable I taking jth value in cth class.
An appropriate number of classes (three) were selected on the basis of the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). The AIC and BIC values of two-, three-, and four-class models is given in the appendix. The values of AIC and BIC were used to develop a three-class model, namely low egalitarian, moderately egalitarian, and highly egalitarian class. In order to arrive at an optimization for achieving the global maxima, up to 300 iterations were fixed in advance and a tolerance in the log likelihood of degree at 10–7 was considered for the convergence of the model. The maximum likelihood ratio test (Chi-square) was attained at the value of 22,783.15 with a p-value < 0.000 (Appendix Table A).
The ‘low egalitarian’ class included men who had a 27% to 56% more chance of justifying hitting or beating wife in five different scenarios and a 13%, 20%, and 24% chance, respectively, of believing that wife was not justified in refusing sex if she knew that the husband had an STI, if the husband had sex with another woman, and if the wife was tired or not in the mood. Men in this class also had the highest chance of giving ‘yes’ as the response to the question if the husband has the justification to get angry or reprimand his wife, refuse financial support, or have sex with another woman if the wife refused sex (Appendix Table B).
In the ‘moderately egalitarian’ class, the chance of justifying hitting or beating wife ranged from 2 to 10%. Moderately egalitarian men had a 67%, 94%, and 89% chance, respectively, of believing that the wife was not justified in refusing sex if she knew that the husband had an STI, if the husband had sex with another woman, and if the wife was not in the mood. The moderate class also included men who had a 2% to 5% chance of giving negative answers to the question on husband being justified in using force, or getting angry, or denying financial support to the wife if she refused to have sex with him (Appendix Table B).
The ‘highly egalitarian’ class included men who had the least chance of giving negative answers to all three batteries of questions. In this category, men had a 0.5% to 5.1% chance of justifying wife beating and a 1.5% to 5.6% chance of believing that the wife was not justified in refusing sex in different situations. They also had a 6% chance of giving ‘yes’ as the answer to the question that husband had the right to get angry if his wife refused sex. For the other questions in this section, they had less than 1% chance of having a negative attitude towards wife refusing sex (Appendix Table B).
For the analysis, bivariate and multivariate techniques were employed. The logistic regression analysis was carried out to understand which factors, among the ones included in NFHS-4 (2015–16), contributed to the covert use of modern reversible contraceptives. In the first model, we included husband’s gender egalitarian attitude variable only to estimate the unadjusted odds ratio of CCU, which gave the pseudo R2 value of 0.011. In the second model, we added couple-level variables and women’s socioeconomic characteristics also and the model produced the pseudo R2 value of 0.083, which shows an improvement over the first model (see chapter 5, Retherford and Choe, 1993, for details) [44].
Prior to the insertion of independent variables into our regression models, we checked the VIF to examine whether there was a multicollinearity among them [45]. The overall value of VIF was 1.4, while the values for the selected independent variables varied between 1.0 and 3.0. This is so because independent variables like place of residence, mobility, working status, wealth status, and freedom to use money by themselves seemed to be not so correlated with each other and with the core variable of interest, that is, husband’s egalitarian gender attitude.