Interviewer effects on underreporting of traditional contraceptive use: evidence from National Family Health Survey-5

1 Background

As of 2021–2023, approximately 65% of in-union women worldwide use some form of contraception, with approximately 59% using modern contraceptive methods (United Nations Population Division, 2024). Globally, approximately 92 million women use traditional contraceptive methods, representing approximately 9%−10% of all contraceptive users. The use of traditional methods tends to be associated with cultural preferences, limited access to modern methods, or concerns about side effects of modern contraception (United Nations Department of Economic and Social Affairs, Population Division, 2022). The overall contraceptive prevalence rate among married women aged 15–49 in India increased from approximately 54% (NFHS-4) to approximately 67% (NFHS-5), reflecting substantial growth in contraceptive use. The modern contraceptive prevalence rate is approximately 56.5% as per NFHS-5, with an estimated 159 million women using modern contraception in 2024. The use of traditional methods has doubled as per the last two rounds of the NFHS survey, increasing from 5.7% in NFHS-4, 2015–2016, to 10.2% in NFHS-5, 2019–2021, along with that 21 states in India show an increase in the use of traditional methods of contraceptives (Prakash et al., 2022). Evidence suggested that in Uttar Pradesh, the use of traditional methods increased by more than 20% over the inter-survey period, a rise considerably higher than national averages (Namasivayam et al., 2023), and this increase is not necessarily due to lack of availability and accessibility, but increasing preferences for traditional methods despite government efforts to make more modern spacing methods available for women (Prakash et al., 2022; Namasivayam et al., 2023). Similarly, in West Bengal and Assam, traditional method use has been reported at levels as high as 17%−18%, and nationally, these traditional methods account for approximately one-third (35%) of spacing contraceptive use (Ram et al., 2014).

The detailed use of contraception in large-scale surveys for women in India is mentioned by NFHS data regularly, using the calendar data method.¹ Despite the widespread and increasing use of contraception, evidence suggests that accurate reporting of contraceptive use, especially for traditional methods, remains a significant challenge in developing countries. In India, traditional contraceptive methods may be particularly prone to underreporting because respondents may not consistently interpret what counts as current use, and reporting in large surveys may be affected by recall and survey-design related errors (Kumar et al., 2022). In developing countries, contraceptive use is often not properly reported, especially for traditional methods, and this happens more often when these methods fail (Polis et al., 2016). It has also been noticed that the highest consistency of contraceptive use reporting was observed among women using sterilization and oral contraceptives (Callahan and Becker, 2012). Past studies have also highlighted that the accuracy of the data collected may be influenced by the interviewer's characteristics, which include their behavior, attitude, and communication style (Cannell et al., 1968). Interviewer bias can affect the respondents' responses, and it will ultimately distort the survey results (Kühne, 2023; Groves, 1989). It is also evident from the research that both the interviewer and respondent play vital roles in the quality of data.

Evidence from existing literature indicates that reporting of contraceptive use varies substantially by method type, with traditional methods being consistently more underreported than modern methods such as sterilization, particularly in reproductive event calendar data (Polis et al., 2016; Rossier et al., 2014). But traditional methods are, and will continue to be, used by women because they are deeply embedded in many cultures and continue to play an important role in pregnancy prevention for millions of women globally (Bertrand et al., 2022). While calendar data are widely used to estimate contraceptive use dynamics (Elkasabi and Assaf, 2021), concerns remain regarding their accuracy, especially for methods that are episodic, culturally sensitive, or socially less acceptable. Parallel research has highlighted that data quality in large-scale surveys is not only influenced by the method of data collection but also by interviewer-related factors, including their socioeconomic background, behavior, communication style, and ability to build rapport with respondents (Cannell et al., 1968; Groves, 1989; Glasner and Van Der Vaart, 2009). Drawing on these two streams of research, this study conceptualizes that greater socioeconomic alignment between interviewers and respondents, such as similarities in marital status, parenthood, place of residence, or shared reproductive experiences, may foster trust and engagement during interviews, thereby improving recall and willingness to disclose reproductive information. Such alignment may be particularly important for reporting traditional contraceptive use, which is more prone to omission and misreporting, particularly in DHS surveys (Rossier et al., 2014; Marston et al., 2017; Rabiu and Rufa'i, 2018; Mairiga et al., 2012). Using reproductive event calendar data from the fifth round of the National Family Health Survey (2022) in India, this study examines how interviewer characteristics contribute to the reporting of traditional contraceptive methods relative to modern methods following non-use of contraception, childbirth, and pregnancy termination, which represent important reproductive transition points at which women are likely to initiate or resume contraception. Building on evidence that later rounds of NFHS have shown improved consistency compared to earlier rounds (Kumar et al., 2022), the present analysis focuses on assessing the quality of NFHS-5 calendar data by investigating the extent to which interviewer characteristics influence reporting accuracy. The overarching objective is to evaluate whether interviewer background plays a critical role in shaping respondents' reporting of traditional contraceptive methods in calendar-based survey data.

2 Methods and materials

2.1 Data sources

For analysis purposes, the National Family Health Survey (NFHS), a periodic and pivotal data set and a nationally representative cross-sectional study overseen by the Ministry of Health and Family Welfare (MoHFW), Government of India, has been used. NFHS meticulously collects data on various dimensions of demographic, health, and socio-economic indicators such as fertility, maternal and child health, reproductive health, infant and child mortality, including contraceptive use, across all states and union territories at the district level in India. NFHS data in India uses a methodology for collecting data on contraceptive use in calendar form, which entails the systematic documentation of contraceptive use over a 60-month time period from the date of interview with the respondent. It provides a detailed and nuanced understanding of reproductive behaviors and reproductive practices (see text footnote 1).

For this study, analysis was conducted on the fifth round of the survey by utilizing the Fieldworker Characteristics file (FW file) and Individual file (IR file). The individual file contains data on women respondents aged 15–49 years, who were interviewed by trained female interviewers in accordance with NFHS survey protocols. The International Institute for Population Sciences (IIPS), which serves as the lead agency for the survey, collaborated with 17 field organizations to conduct data collection. The survey collected detailed information from 636,699 households, involving 724,115 women across 707 districts nationwide during the period 2019–2021, achieving an individual response rate of 97%. It also included information about the interviewer who collected the survey data. A two-stage sample design was used with probability proportional to size to select the primary sampling units (PSUs) at the first stage, where villages served as primary sampling units (PSUs) for rural areas, and census enumeration blocks (CEBs) were selected from urban areas. In the second stage, a fixed number of households (22) were systematically sampled from the selected PSUs (see text footnote 1). The sampling and data collection procedure is well documented on the DHS program web portal (see text footnote 1). Out of 7,127, duplicate fieldworker entries from 2010 were removed, and ultimately only 2,950 female interviewers were retained in the final dataset for the analysis. The FW file contains information about interviewer characteristics collected via a self-administered questionnaire at the outset of fieldwork. Upon completion of training, each interviewer was assigned a unique numeric identification code (ID), which facilitated linking fieldworker data with respondent data. During data cleaning for the study purpose, duplicate interviewer IDs were identified because, in some cases, replacement interviewers were assigned the same ID when the original interviewer left the survey. These duplicate entries were removed to ensure a one-to-one linkage between interviewer and respondent data.

2.2 Outcome variable

The contraceptive calendar recorded month-by-month information in a string format on all reproductive events for the last 5 years prior to the survey (60 months). From this contraceptive calendar information, the outcome variable is generated by categorizing the women into two groups: women who did not report opting for a traditional contraceptive method following periods of non-use of contraception, childbirth or pregnancy termination, and those women who reported opting for a traditional contraceptive method after following periods of non-use of contraception, childbirth, or pregnancy termination. Female sterilization, male sterilization, emergency contraception, and missing were dropped from the study, as previous studies have reported high consistency in sterilization reporting (Callahan and Becker, 2012).

2.3 Explanatory variables

In this study, we have examined various demographic characteristics of interviewers to understand their potential influence, primarily on contraceptive calendar data. These included age in completed years categorized into four groups (15–20, 21–25, 26–30, 30 and above) as well as interviewer age relative to women respondent coded as younger than, same as, older than the respondent, education level also used in two groups (primary, secondary, higher) and interviewer education also categorized as relative to women respondent (same and different), religion (Hindu, Muslim, other), religion also used as same or different relative to women respondent, caste (ST, SC, OBC, other), caste also used as same or different relative to women respondent, previous work experience in NFHS/ Demographic Health Survey (DHS), and any other surveys along with variable as employment by agency (yes, no), marital status (ever married, never married), children ever born (no children, have children) and place of residence (rural, urban). Along with that, another variable was created to account for the total number of interviews done by the interviewer in the survey period (low with a mean of 16.6 interviews, medium with a mean of 28.4 interviews, and high with a mean of 41.3 interviews). These analyses aim to uncover any potential biases or influences interviewer characteristics may have on survey responses, particularly concerning sensitive topics such as family planning methods.

2.4 Control variables

To ensure a comprehensive examination, the study also adjusted for several potential confounders. As the study was about contraceptive use pattern and reporting, it was imperative to include variables relevant to women: age of women respondents in completed years (15–20, 21–25, 26–30, 30 and above), current marital status (never married, ever married), wealth quintile (poorest, poor, middle, rich, richest), place of residence (rural, urban), caste of respondents (Scheduled Tribes/Scheduled Castes, Other Backward Classes, [including None of them and Don't know]), religion (categorized as Hindu, Muslim, [including Christians, Sikhs, Buddhists/Neo-Buddhists, Jains, Jews, Parsis/Zoroastrians, individuals reporting no religion, and other religions]). By including the above variables as control variables in this study, the study aimed to diminish the confounding effect and enhance the research findings.

2.5 Methodology

2.5.1 Statistical analysis

Analysis was performed using STATA 16 and MLwiN Version 3.05. Descriptive statistics, coupled with bivariate analysis, were used to explore the distribution of interviewer characteristics. The socioeconomic and demographic profiles of interviewers were presented using descriptive analysis. Non-response among women was adjusted by applying appropriate sampling weights during analysis. Moreover, a multilevel cross-classified model analysis was utilized to address the objectives of the study. The findings were conveyed through adjusted odds ratios (AORs), accompanied by a 95% confidence interval (CI). Furthermore, the intra-class correlation coefficient (ICC) was calculated from a cross-classified random intercept multivariable four-level logit model, which was used to assess the effect of interviewers on the above-defined outcome variables while adjusting for all the independent variables mentioned above.

2.5.2 Cross-classified multilevel model

A four-level, cross-classified, random intercept multivariable logit model was used to examine interviewer effect on the outcome variables. The result for the ith respondent—interviewed by the jth interviewer at the kth PSU in the lth district—was specified in this model. Each interviewer worked for numerous PSUs and districts, and each PSU and district used multiple interviewers, resulting in a cross-classification of the random intercept model. Unlike classic PSU-district models, this system does not have a perfectly hierarchical structure. To disentangle the micro-macro interactions when the contextual level does not present a clear or “natural” structure, we used a cross-classified model. This approach allows for more flexible specifications, making it a valuable tool for analyzing complex, non-hierarchical data relationships (Zaccarin and Rivellini, 2002). The following flowchart shows the hierarchical structure.

Socioeconomic and demographic factors at the individual, fieldworker, and PSU levels comprise the independent variable. The model is defined as:

log(πi(j,kl)1-πi(j,kl))=β0+β1Txi(j,kl)+β2Tx0k+ ε0j+ε0k+ε0l

In this case, π_{i(j, kl)} denotes the likelihood that, for one of the dependent variable categories as previously described, the ith respondent questioned by the jth interviewer in the kth PSU of the lth district will get the value of 1. The coefficients of socioeconomic and demographic factors at the person level are contained in β1T vector, while the coefficients of covariates at the PSU level are contained in β2T vector. The error terms at the interviewer, PSU, and district levels are denoted by the symbols ε_0j, ε_0k, and ε₀₁, in that order.

ε0j~N(0,σ0j2),ε0k~N(0,σ0k2),ε0j~N(0,σ0l2) ;

where σ² are variances at different levels.

The Markov Chain Monte Carlo (MCMC) estimation process was used to estimate the cross-classified random intercept models, and the model was executed using the RunMLwiN program in STATA 16.0. Appropriate sampling weights were applied in the analysis to ensure representative estimates. In this study, ICC was calculated, which represents the share of the total variance in the outcome variable at all levels. As the study used the logit link function in the cross-classified model analysis, the individual-level variance is a fixed value, i.e., 3.29, and has been used in NFHS data (Singh et al., 2024). Figure 1 shows the flowchart of the hierarchical structure:

At the district level, ICC calculated the relationship between two distinct women living in the same district but being questioned by several field workers at various PSUs.

ICC particular level=selected level variance (District level variance+PSU level variance+Investigator level variance+individual level variance)ICC at FW level=Fieldworker level variance (District level variance+PSU level variance+Investigator level variance+individual level variance)=[σ0j2σ0l2+σ0k2+σ0i2+3.29]ICC at District level=District level variance (District level variance+PSU level variance+Investigator level variance+individual level variance)=[σ0l2σ0l2+σ0k2+σ0j2+3.29]ICC at PSU level=PSU level variance (District level variance+PSU level variance+Investigator level variance+individual level variance)= [σ0k2σ0l2+σ0k2+σ0j2+3.29]

The ICC at each level represents the correlation in the outcome between two observations sharing the same higher-level unit. The interviewer-level ICC reflects similarity between observations handled by the same interviewer, the PSU-level ICC reflects similarity between observations within the same PSU but handled by different interviewers, and the district-level ICC reflects similarity between observations within the same district but belonging to different PSUs and handled by different interviewers.

3 Result

3.1 Profile of the Interviewer, NFHS-5

The data presented in Table 1 outline the demographic profile of interviewers involved in the study. Of the total 2,950 investigators, 43% hailed from urban areas. A majority of interviewers, approximately 61%, fell within the age range of 21–25 years, with an average age of 24.76 years (standard deviation = 4.42 years). The vast majority of interviewers, accounting for 87%, had never been married, and 92% had no children. A majority of interviewers (89%) are highly educated. In terms of religious affiliation, the majority of interviewers identified themselves as Hindu, comprising 73% of the total, followed by Christianity (11%). Regarding caste distribution, approximately 31% belonged to other backward classes (OBCs), 25% were from scheduled castes (SCs), and 23% were from scheduled tribes (STs). Only 22% of interviewers had previous experience working in NFHS, and 28% had participated in surveys other than NFHS in the past. Notably, around a quarter (24%) of the interviewees were currently hired by an agency at the time of their recruitment for NFHS. Approximately 28% of the interviewers are Hindi speakers.

Table 1. Profile of interviewers, NFHS-5 (2019–2021).

3.2 Fieldworker-respondent duo attributes

In Table 2, a comparative analysis of interviewers' characteristics concerning residence, marital status, age, religion, and caste in relation to women respondents reveals notable differences and commonalities. Specifically, 13.8% of field investigators had a different state of residence compared to respondents, while 43.9% of fieldworkers had a different type of place of residence than respondents. 12.5% of fieldworkers differed in marital status from respondents. Age-wise, 75.1% of investigators were younger than respondents. Regarding religion, 70.9% of interviewers and respondents belonged to the same religion, and only 36.7% of interviewers and respondents had the same caste. Approximately 83.3% of respondents and their interviewers spoke the same native language. About equal percentages of interviewers conducted a low, moderate, and high number of interviews, respectively. These findings provide insights into demographic dynamics between investigators and respondents.

Table 2. Fieldworker-respondent duo attributes, NFHS-5 (2019–2021).

3.3 Variance at interviewer level: findings from a cross-classified four-level multilevel model

Table 3 displays the estimated district, PSU, and interviewer-level intraclass correlation coefficients for reporting opting for traditional contraceptives after no contraceptive method use, childbirth, or pregnancy termination. These outcomes were derived from a four-level random intercept multivariable logistic regression model. Following the adjustment of respondents' and interviewers' background variables, detailed estimates of each of the outcome variables and their constituent domains are provided in Table 3. The analysis has found a significant impact of the characteristics of interviewers on the reporting of opting for traditional contraceptives after women have had a birth, pregnancy termination, and non-use of contraception. The interviewer effect for the outcome variable indicates that approximately 19% of the variability in the outcome variable is attributable to differences between districts, approximately 13% of the variability in the outcome variable is due to differences between PSUs, and approximately 29% of the variability in the outcome variable is due to interviewers.

Table 3. Outlining the percentages of variance attribution for reporting of traditional contraceptive method use after childbirth, pregnancy termination, and no method use, at district, PSU, and interviewer levels using the cross-classified four-level model NFHS-5, 2019–2021.

3.4 Percentage of women reporting traditional contraceptive use after birth, pregnancy termination, and no method use by interviewer characteristics

The highest reporting was observed among interviewers aged 21–25 years and those aged 30 and older. Interviewers with secondary education recorded the lowest reporting of traditional method use (19.60%) compared with other interviewers.

Reporting of traditional contraceptive use among women by the interviewer revealed notable variation across interviewer characteristics. Interviewers with secondary education recorded lower reporting of traditional contraceptive use (20%) as compared to interviewers who completed higher education and above (33%). The findings indicate limited variation in reporting by the place of residence of the interviewer and whether the interviewer had children or not.

Reporting was slightly higher among interviewers who were employed by the agency than among those who were not.

Reporting was relatively lower when the interviewer and respondent had different marital statuses (30%), whereas it was higher when both shared the same marital status (33%). When the interviewer was younger than the respondent, the reporting of traditional method use was relatively higher than that of women of the same and higher age than the interviewer. Reporting of traditional contraceptive use is significantly varied by interviewers' religion, with higher reporting when religion is the same.

3.5 Multilevel cross-classified adjusted model, for reporting of traditional contraceptive use after women have had a birth, pregnancy termination, and no method use

Table 4 shows the results obtained from bivariate analysis and multilevel cross-classified model for women who reported opting for traditional contraceptives following birth, pregnancy termination, and no method use, by interviewer characteristics. Interviewers aged 30 years and above showed a positive effect, with women interviewed by those aged 30 years or older being more likely to report traditional contraceptive use (adjusted odds ratio (AOR) = 1.40, p < 0.01), compared to the younger group (15–20 years), while no significant differences were observed for the 21–25 and 26–30 groups. The religion of the interviewer was associated with differences in reporting traditional contraceptive use. Compared to Hindu interviewers, Muslim interviewers were associated with significantly lower reporting (AOR = 0.43, p < 0.01). Caste background is also influenced by reporting. Women interviewed by interviewers from OBC backgrounds had a significantly lower likelihood of reporting traditional contraceptive use compared to those interviewed by non-OBC interviewers (AOR = 0.86, p < 0.01). Women interviewed by interviewers with secondary education were less likely to report traditional contraceptive use compared with those interviewed by primary-educated interviewers (AOR = 0.54, p < 0.01), while no significant difference was observed for interviewers with higher education. Previous experience in NFHS (DHS) surveys or past employment with a survey agency was not significantly associated with reporting. However, interviewers without experience of other surveys were less likely to report traditional contraceptive use (AOR = 0.90, p < 0.10).

Table 4. Percentage and multilevel cross-classified adjusted odds ratio of women reporting opting for traditional contraceptives following birth, pregnancy termination, and no method used by interviewer characteristics, India, NFHS, 2019–2020.

^*p < 0.05,

^**p < 0.01,

^***p < 0.001.

Interviewer proficiency in administering the NFHS-5 schedule, particularly the contraceptive calendar, had a significant positive effect on reporting. Women interviewed by those who conducted a moderate number of interviews (AOR = 1.35, p < 0.01) or a high number of interviews (AOR = 1.26, p < 0.05) were more likely to report traditional contraceptive use than those interviewed by less experienced interviewers.

Interviewer–respondent similar characteristics also influenced reporting. Women were less likely to report traditional contraceptive use when the interviewer was of a similar age (AOR = 0.89, p < 0.05) or older (AOR = 0.86, p < 0.01) than the respondent. Interviewers from the same state as the respondent were associated with lower reporting (AOR = 0.82, p < 0.01). In contrast, women were more likely to report traditional contraceptive use when the interviewer and respondent shared the same marital status (AOR = 1.42, p < 0.01) or religion (AOR = 1.21, p < 0.01). However, sharing the same caste was associated with slightly lower reporting (AOR = 0.94, p < 0.05).

4 Discussion and conclusion

This study highlights an often overlooked issue in contraceptive data collection, that is, the substantial underreporting of traditional contraceptive methods in the NFHS-5 contraceptive calendar data. Understanding these discrepancies requires careful consideration of the role played by interviewers during data collection. The present study examined the underreporting of traditional contraceptive methods through the lens of interviewer characteristics and their influence on the reporting of traditional contraceptive use following birth, pregnancy termination, and periods of non-use.

The findings reveal that interviewer characteristics, including age, educational attainment, prior survey experience, and workload, can significantly influence the accuracy of the data collected. Furthermore, similarities and differences between interviewers and respondents in characteristics such as state of residence, marital status, age, religion, and caste were found to affect the quality and completeness of contraceptive reporting.

Interviewer education exhibited a U-shaped association with reporting patterns. Women interviewed by secondary educated interviewers were less likely to report traditional contraceptive use, whereas no significant difference was observed among those interviewed by highly educated interviewers. One possible explanation is that interviewers with primary education may establish stronger rapport with respondents, particularly in rural settings, thereby encouraging more open disclosure of culturally familiar contraceptive practices. In contrast, secondary-educated interviewers may probe differently, and it may create discomfort among respondents because of perceived educational differences. Highly educated interviewers, on the other hand, may maintain greater professional neutrality, thereby avoiding both the advantages of increased rapport and the disadvantages associated with social distance. Previous research has demonstrated that interviewer-respondent educational matching can influence survey participation and response behavior (Durrant et al., 2010). Studies have further shown that educational similarity often leads to higher cooperation rates, while respondents with lower educational attainment may provide more substantive responses when interviewed by college-educated interviewers (Yang and Yu, 2008). Together, these findings suggest that educational differences between interviewers and respondents can shape disclosure patterns in complex ways.

Interviewers aged over 30 years were associated with higher reporting of traditional contraceptive use. When interviewers and respondents belong to the same state of residence, the resulting social proximity appeared to have a negative effect on reporting. In close-knit communities, respondents may fear breaches of confidentiality, which can discourage the disclosure of sensitive information. This finding is consistent with earlier research by Anglewicz et al. (2019), who noted that many large-scale surveys deliberately recruit interviewers from outside the study community to minimize concerns about confidentiality. Nevertheless, their studies also demonstrated that interviewer-respondent familiarity can significantly influence the reporting of sensitive behaviors, including traditional contraceptive use and unwanted births. Rural respondents, who are more likely than urban respondents to know local interviewers personally, often exhibit higher levels of honest disclosure when interviewed by someone from a similar rural background (Anglewicz et al., 2019).

The study also found that interviewer-respondent similarity in marital status positively influenced the reporting of traditional contraceptive use. This finding aligns with previous evidence showing that respondents are more likely to cooperate and engage meaningfully in surveys when interviewed by individuals who share similar demographic characteristics (Webster, 1996).

With respect to religion, interviewers belonging to the Muslim religion were negatively associated with the reporting of traditional contraceptive use, suggesting that interviewer-respondent religious dynamics may influence reporting behavior. The observed variation may reflect interviewer-respondent interaction dynamics rather than inherent religious characteristics. At the same time, interviewer-respondent religious concordance was associated with higher reported, suggesting that a shared religious identity may create trust and a greater sense of comfort during the interview process.

In contrast, interviewer-respondent similarity in caste was associated with lower reporting of traditional contraceptive use. This may be attributed to concerns about social judgment or belief of a breach of confidentiality within caste-based social networks, where privacy cannot be maintained. The contrasting effects of religious and caste concordance indicate that shared identity does not uniformly promote disclosure; rather, its influence depends on the social meaning and stigma attached to specific identities and behaviors.

Although it may be assumed that matching interviewers and respondents based on demographic characteristics would improve data quality, the existing evidence provides limited support for this assumption. In some cases, deliberate matching may even introduce new sources of bias to survey responses (Davis et al., 2010). Previous research has demonstrated that interviewer-respondent acquaintance is associated with a range of family planning outcomes, highlighting the need for careful consideration of interviewer recruitment and deployment strategies in survey design (Anglewicz et al., 2019). Furthermore, studies have shown that collecting contraceptive calendar data requires substantially more time than administering conventional questionnaires. Calendar-based data collection has been reported to take approximately twice as long as traditional survey approaches (Engel et al., 2001) and, when recall aids are used, approximately 12 min longer on average than standard questionnaire-based interviews (van der Vaart, 1996).

Cultural barriers may further limit open discussion of sensitive reproductive health topics, particularly when traditional contraceptive methods are practiced privately and remain outside formal health care systems. In such settings, respondents may be reluctant to disclose their contraceptive behaviors during interviews, especially when other household members are present, thereby contributing to underreporting (Malik et al., 2023). Older interviewers may be perceived as more experienced and trustworthy, which could facilitate respondent cooperation. However, age differences between interviewers and respondents may also influence reporting behavior. Younger women, for example, may feel less comfortable discussing intimate reproductive matters with substantially older interviewers whom they perceive as moral elders. Such generational differences may reduce willingness to disclose privately practiced contraceptive methods. Previous studies have similarly reported that younger respondents may underreport sensitive behaviors when interviewed by older interviewers because of concerns about judgment or social desirability (Rozelle et al., 2023).

Research has also shown that older respondents often experience greater difficulty understanding survey questions than younger respondents. Interestingly, this challenge becomes substantially greater when the interviewer is considerably younger, increasing by approximately 3-fold compared with interviews conducted by older interviewers (Życzyńska-Ciołek and Kołczyńska, 2021). The lower reporting of traditional contraceptive methods among respondents interviewed by Muslim interviewers may also reflect broader religious norms and beliefs that discourage contraceptive use in certain contexts (Alomair et al., 2023; Shumayla and Kapoor, 2017). Previous studies have highlighted similar interviewer effects. For example, in Egypt, Muslim women reported higher levels of religiosity when interviewed by veiled women (Blaydes and Gillum, 2013). Similarly, in Morocco, religious respondents tended to provide more pious responses to interviewers wearing a hijab and less religious responses to interviewers with a more neutral appearance. These findings suggest that respondents may modify their responses to align with perceived expectations, social norms, and cultural dynamics associated with an interviewer's religious identity or appearance (Benstead, 2014).

Interviewer experience is another important determinant of data quality. Experienced interviewers are generally better equipped to manage complex and sensitive topics and are therefore more likely to collect consistent and reliable data. Interviewer effects are particularly pronounced in reproductive health surveys that require detailed recall, such as questions regarding contraceptive use and duration (Elksabi and Khan, 2022). Research shows that interviewers usually become faster at conducting interviews as they gain more experience. Compared to their first interview, later interviews often take less time because interviewers become more familiar with questionnaire and survey procedures. However, being faster may also mean that interviewers ask questions more quickly, spend less time explaining questions, or do less probing for additional information. As a result, respondents may provide less detailed answers, which could affect the quality and accuracy of the data collected (Olson and Peytchev, 2007). These findings underscore that the underreporting of traditional contraceptive use is not solely a consequence of respondent recall limitations or reluctance to disclose information. Rather, it is also shaped by the extent to which interviewers are able to establish rapport, foster trust, and effectively probe for complete responses. The results highlight the importance of considering both interviewer and respondent characteristics when designing and implementing large-scale surveys.

From a policy perspective, these findings suggest that the NFHS-5 contraceptive calendar may underreport the usage of traditional contraceptive methods, potentially leading to inaccurate assessments of family planning behaviors and needs. Such underreporting may distort estimates of contraceptive prevalence, unmet need for family planning, method satisfaction, and contraceptive continuation rates, thereby affecting program planning and service delivery. To improve data quality, interviewer recruitment and deployment strategies should take into account contextual and demographic factors that may influence respondent disclosure. In addition, interviewer training may benefit from being extended beyond technical instruction on contraceptive calendar administration to include communication skills and rapport-building techniques that may facilitate respondent disclosure. However, the effectiveness of such training requires further empirical evaluation.

Despite its contributions, this study has several limitations. Although multilevel cross-classified models were used to account for clustering at the interviewer level, unmeasured factors such as respondents' attitudes and the degree of privacy during interviews may still have influenced the findings. Furthermore, the contraceptive calendar relies on retrospective self-reported information and is therefore susceptible to recall bias. It is also possible that some respondents do not perceive traditional methods as legitimate contraceptive practices and therefore fail to report them, even when interviewers make efforts to obtain complete information.

Future research should further investigate how underreporting influences our understanding of contraceptive dynamics over time and the implications for demographic and public health research. Particular attention should be given to the role of interviewer-respondent characteristics in shaping the reporting of sensitive and frequently underreported reproductive health behaviors and outcomes.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Generative AI was used for language editing and manuscript refinement. All analytical work and interpretations are the author's own.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Sunil Raina, Dr. Rajendra Prasad Government Medical College, India

Reviewed by: Michael Bailey, Georgetown University, United States

Kassu Mehari Beyene, Barrow Neurological Institute (BNI), United States

Footnotes

PSU, primary sampling unit; ICC, intra-class correlation coefficient; NFHS-5, National Family Health Survey-5; DHS, Demographic and Health Surveys.

Footnotes

https://www.nfhsiips.in/nfhsuser/index.php (Accessed February 5, 2026).