METHODS

Survey methods

Natsal 1990 was carried out in 1990–1 and the methods have been described in detail elsewhere.¹⁸ Natsal 2000 was carried out in 1999–2001.⁷ In both surveys a multistage probability sample of addresses was drawn from the postcode address file for Britain, and hence an appreciable change in the effect of sampling frame exclusions on bias is unlikely. In Natsal 2000 London addresses were oversampled.⁷ One member of the household was randomly selected for interview, without replacement. In Natsal 1990 those aged 16–59 were eligible for interview, while in Natsal 2000 this range was reduced to 16–44. In both surveys similar information was collected by the interviewer in a face to face interview, but the more sensitive information was collected through a self completion method (PAPI in Natsal 1990 and CASI in Natsal 2000). Those with minimal sexual experience were not offered the self completion section.⁷ The questions in Natsal 2000 were substantially the same as those in Natsal 1990, but additional questions were also included to reflect changing policy and epidemiological interests. A total of 13 765 interviews were collected in Natsal 1990 among those aged 16–44 and 11 161 in Natsal 2000.

Methods of exploring bias and its change between surveys

To measure the representativeness of the sample of participants we compared the response rates in the two surveys, and the extent to which the demographic structure of the participants (after weighting for differential selection probabilities, but not non-response) matched the contemporary population, as indicated by population estimates for mid-1991¹⁹ and mid-1999.^20,21

As an external consistency check, we compared the reporting of abortion in the previous year with contemporary official statistics from 1990^22,23 and 1999 (provided by the Office for National Statistics).

We analysed data from the 19 year common birth cohort, those eligible for both surveys, aged 16–34 in Natsal 1990 (3857 male and 5161 female respondents) and 26–44 in Natsal 2000 (3392 male and 4747 female respondents). We compared the reporting of events occurring before 1990 between the samples interviewed in Natsal 1990 and Natsal 2000. We would expect little difference if there is no change in bias. To establish that the experience occurred before 1990, age at first occurrence is required, which can be compared with age at interview. Events reported in Natsal 1990 are taken to have occurred before 1990, and reporting in Natsal 2000 of an event at age 10 or more years less than current age is taken to have occurred before 1990. Heterosexual intercourse before 16 and homosexual experience were included, and less sensitive experiences—for example, not living with both parents until 16, were included for comparison. These experiences were selected to cover the full range of objective questions available relating to specific ages. We looked for evidence of variation in the change in any bias according to sex by testing its interaction with survey. This method assumes either a broadly closed population over time—that is, minimal emigration and immigration, or that immigrants and emigrants have comparable sexual behaviour.

For each survey we determined the extent to which reporting of homosexual experience varied between the self completion and face to face components of the interview.

We compared the reporting of homosexual genital contact before 20 (heterosexual intercourse before 16 has been examined elsewhere⁸), across age groups in Natsal 2000. This analysis was performed not to address the change in bias between surveys directly, but to assess whether evidence from Natsal 2000 alone can demonstrate substantial change in behaviour over the past decade.

We compared the reporting of attitudes to sex before marriage, unfaithfulness, casual sex, and male homosexual sex. Odds ratios greater than 1 presented for the reporting in Natsal 2000 relative to Natsal 1990 indicate increased tolerance.

We compared the proportions of respondents reporting other people present in the home, and other people within earshot, in the two surveys. This could be an important factor determining willingness to report, although both surveys were designed so that participants were not required to give verbal answers to any sensitive questions.

Statistical methods

Our initial comparison of the demographic structure of the participants with population estimates was performed using data weighted only for differential selection probabilities, as defined by household size and region. For all other analysis, the data were weighted to account for the differential selection probabilities and then also post-stratified to British population estimates of the age, sex, and region distribution from mid-1991 and mid-1999 as described elsewhere.⁷ Post-stratification is particularly intended to reduce any bias from non-participation. All analysis was performed using the survey analysis software of stata 6, which accounts for the clustering, weighting, and stratification of the data. We used the odds ratio (OR) for Natsal 2000 relative to Natsal 1990 as the measure of change in reporting of behaviours. For ordinal outcomes (for example, attitudes), these are calculated under the assumption of proportional odds. The adjusted ORs presented are calculated using logistic regression.

RESULTS

The participation rates (after adjusting for the oversampling of London in Natsal 2000, and with other standard adjustments⁷) were similar at 66.8% in Natsal 1990 and 65.4% in Natsal 2000. However, these rates are not strictly comparable since the age ranges eligible were 16–59 in Natsal 1990 but 16–44 in Natsal 2000. Since people over 45 were found less likely to participate in Natsal 1990² it may be that the fall in participation rate between surveys would have been somewhat larger had the age range remained constant. Table 1 shows that in Natsal 1990 men were under-represented, and in Natsal 2000 this under-representation is somewhat greater. Among men the under-representation of London is somewhat increased, among women the under-representation of those aged less than 30 is somewhat increased. Among eligible participants who accepted the self completion component of the interview the non-response rates for the questions within were generally very low using PAPI (Natsal 1990) but even lower using CASI (Natsal 2000) owing in part to the automatic routing of questions, as found previously¹⁵ (data not shown).

In Natsal 1990, the reported abortion rate among the 7384 female respondents aged 16–44 was 12.9/1000. This is 84% (95% CI 66 to 107) of the rate from official statistics. In Natsal 2000 the reported rate based on 6173 respondents was 13.5/1000. This is 86% (67 to 111) of the official rate, and hence the completeness of reporting seems largely unchanged, providing no evidence for a change in bias. For each survey the completeness of reporting was also calculated by age group (16–19, 20–24, 25–29, 30–34, 35–39, and 40–44), data not shown. For neither survey was the reporting for any age group significantly different from the official rate, and in neither survey was there evidence of a trend in completeness across age groups.

Table 2 shows the prevalence of reported behaviours in each survey for the common birth cohort and the corresponding ORs. The reporting of both homosexual experience and heterosexual intercourse before 16 are significantly higher in Natsal 2000 for women with ORs (95% CI) of 2.00 (1.61 to 2.48) and 1.49 (1.31 to 1.69), respectively. The reporting is also significantly higher for men, but less than for women, with ORs of 1.80 (1.46 to 2.21) and 1.15 (1.02 to 1.29), respectively. The increase in reporting of heterosexual intercourse before 16 is significantly less for men than for women. The reporting of the other experiences, which are generally somewhat less sensitive, is broadly unchanged with the exception of significantly higher reporting in Natsal 2000 of not discussing sexual matters with either parent when aged 14.

In Natsal 1990, among male respondents 5.3% (95% CI 4.7 to 5.9) reported homosexual experience in the face to face component of the interview, compared to 6.0% (5.4 to 6.7) in the self completion component, a ratio of 1.14. Among women the figures were 2.8% (2.5 to 3.3) and 3.7% (3.3 to 4.2), a ratio of 1.30. In Natsal 2000 among men the figures were 6.7% (6.0 to 7.5) and 8.4% (7.6 to 9.2), giving a ratio of 1.24, and among women the figures were 7.0% (6.4 to 7.8) and 9.7% (8.9 to 10.5) giving a ratio of 1.38. Hence, among men and women the reporting of homosexual experience in self completion relative to face to face was somewhat higher in Natsal 2000 (when CASI was used) than Natsal 1990 (when PAPI was used).

The reported prevalence of homosexual genital contact before 20 among men in Natsal 2000 in the age groups 20–24, 25–29, 30–34, 35–39, and 40–44 were 3.6% (2.3 to 5.5), 3.1% (2.1 to 4.5), 5.6% (4.2 to 7.4), 5.2% (3.9 to 6.9), and 4.6% (3.3 to 6.3), respectively. Among women the figures were 3.9% (2.7 to 5.6), 2.6% (1.7 to 3.8), 2.2% (1.5 to 3.2), 2.3% (1.6 to 3.4), and 2.6% (1.8 to 3.8). For neither sex was there a significant trend in the prevalence across age. As reported elsewhere,⁸ reporting of heterosexual intercourse before 16 in Natsal 2000 is greater among younger women, but not among men. This suggests that this behaviour has increased among the population of women over recent years (although the results could in principle reflect a reduced willingness to report the behaviour among older respondents).

Table 3 shows marked changes in reported attitudes between Natsal 1990 and 2000. There are substantial increases in the reported tolerance of male homosexuality OR 2.10 (1.93 to 2.29) among men and 2.95 (2.74 to 3.18) among women, and of “one night stands” and, in addition, the reported tolerance of sex before marriage is increased among women. Testing of interaction terms (data not shown) indicates that the increase in reported tolerance is significantly smaller for men. The reported tolerance of unfaithfulness in marriage or cohabitation is significantly decreased in Natsal 2000, though tolerance was already very low in Natsal 1990.

The proportion of interviews where someone else was in the home but not within earshot, was slightly higher in Natsal 2000 than in Natsal 1990, at 30.6% (29.6 to 31.6) compared with 26.2% (25.3 to 27.1). However, the proportion of interviews where someone else was within earshot was much lower at 28.0% (27.0 to 29.0) compared to 45.1% (44.1 to 46.1).

CONCLUSIONS

The comparison of data from Natsal 1990 and 2000 demonstrates significant increases in the reported prevalence of several sexual behaviours among those aged 16–44.^6,7 Such increases may reflect real increases in the population prevalence, but they might reflect in part a change in bias. In this paper we have attempted to assess whether there has been an appreciable change in bias and, if so, whether it has led to higher or lower reporting of sensitive behaviour, so that inference about real changes over time in population behaviour can be made from the surveys.

The observed increase in tolerance towards sexual diversity reflecting changed social norms, the higher proportion of interviews conducted without other people within earshot, and the use of CASI interviewing in Natsal 2000 would be expected to have led to more candid responses. These changes may also have led to greater participation in the survey among those with sensitive sexual experience. Both these changes in bias are consistent with the results of the analysis of the common birth cohort, those eligible for both surveys. The reporting of homosexual experience before 1990 is almost twice as high in Natsal 2000 as Natsal 1990, and the reporting of heterosexual intercourse before 16 is also significantly higher. For the other (generally less sensitive) experiences, with only one exception there is no evidence of a change in reporting between surveys. The finding that in Natsal 2000 the reporting of homosexual experience in the self completion relative to the face to face component was higher than in Natsal 1990 is consistent with a small increase in the candour of reporting because of CASI. The comparison of the demographic structure of the two samples, after weighting for differential selection probabilities, with contemporary population estimates demonstrated small changes. Finally, the comparison of prevalence estimates for recent abortion with contemporary statistics provided no evidence of an appreciable change in bias. So while not every method suggests the same conclusion, the evidence taken together is consistent with a change in bias, leading to higher reporting of sensitive behaviours in Natsal 2000. We cannot deduce that the sample of participants is more representative in Natsal 2000, but the evidence suggests that reporting in Natsal 2000 is more candid. This finding is clearly encouraging, and since also CASI interviewing leads to higher internal consistency and lower question non-response,¹⁵ we are hopeful that data from Natsal 2000 have not only updated population estimates but improved their accuracy.

Clearly our work has limitations. Only a few behaviours can be considered in each of our methods, and for the common cohort analysis these cannot relate to recent experience. Only abortion statistics are available for reliable consistency checks external to the surveys. It is unclear how our findings can be generalised to other experiences. The interpretation of the results from the common birth cohort analysis is also not straightforward. The higher reporting observed in Natsal 2000 for sensitive behaviours may be in part because when interviewed, those respondents from Natsal 2000 were 10 years older than those in Natsal 1990 and so a longer period of time had elapsed since the events reported. Difficulties of recall are unlikely to have led to such large differences in reporting, because these are personally significant events, and the results for the less sensitive events show no evidence of systematic problems. However, willingness to report the events may have increased with the time elapsed, as the event became less morally charged.

To make inference concerning change in population behaviour from the surveys, for some behaviours data from external sources—for example, abortion statistics, can be used to aid interpretation. The change in early experience can be established using data from Natsal 2000 alone. For other sensitive behaviours where change must be established by a comparison of data from the two surveys, the likelihood of a change in bias must be considered. Any change in bias could be highly variable across behaviours—for example, being large for early homosexual experience but minimal for recent abortion, and might vary by sex. Thus, increases in the reported prevalence over time, and greater increases for women, may overestimate the real change in population behaviour. However, looking at Natsal 2000 data alone there is evidence of real increases in the prevalence of early heterosexual intercourse among women (though little evidence of a change in early homosexual experience for either sex). The comparison of reported attitudes between the surveys and other sources suggests greater tolerance of sexual diversity which arguably itself provides evidence of population behaviour change. Furthermore, external data such as the recent increase in STI incidence,⁴ suggest that the increases in the reporting of various sensitive behaviours associated with STI risk between surveys⁷ in part reflect changing population behaviour, and cannot be explained as merely the result of changing bias.

CONTRIBUTORS

AC was the lead writer of this paper; AC and CM performed the statistical analysis; CM prepared the dataset for analysis with contributions from CK and AC; AJ, KW, BE, and KF were the principal investigators and designed and managed Natsal 2000, with contributions from AC, SM, and CK; WM, KN, and all the authors contributed towards the ideas for analysis, interpretation of the results, and writing of the paper.