A cost-benefit analysis of a condom social marketing programme in Tanzania

Abstract

This article uses the revealed preference, willingness-to-pay approach to estimate the benefits in a cost-benefit analysis of a condom social marketing (CSM) programme in Tanzania. The demand curve used to derive the consumer surplus had unit elasticity and it was estimated from a cross-sectional sample of 1272 persons. People were willing to pay different prices for the condoms because perceived quality varied. Net benefits were close to zero for the minimum estimate that ignored external benefits. With external benefits included, the CSM programme was judged socially worthwhile with our best estimate producing a benefit-cost ratio ranging from 1.31 to 1.72.

Notes

¹On CBA criteria generally, see Brent (1996, ch. 1; 1998, ch. 2); and for CBA criteria applied exclusively to health care evaluations, see Brent (2003).

²For a CBA of female primary education as a means of reducing HIV/AIDS in Tanzania using an alternative benefit methodology to WTP (i.e. the human capital approach) see Brent (2007).

³An alternative, revealed preference, WTP methodology utilized in CBAs of HIV/ADS involves the statistical life approach, see McKay and Phillips (1991) and La Croix and Russo (1996). This approach uses labour market choice data which relates to a trade-off of wages for differences in health risks. The main problem with this method is that individuals find it difficult to perceive small risk differences and so the trade-off estimates could be unreliable. Also, the method assumes that the trade-off is linear. For a full discussion see Brent (2007, ch. 8).

⁴A parallel programme was run by the government that supplied free condoms. Although 37 respondents in the PSI survey did report a zero purchase price for their condoms, we exclude these observations because the survey was directed at Salama users. Respondents were consumers leaving condom outlets rather than health clinics, where the government condoms were mainly supplied.

⁵There were three main reasons why a price was not recorded for all 2533 respondents: (a) The price was missing. In which case, the respondent did not know the answer, or did not want to give an answer. (b) The price was zero. As explained in Footnote 3, the government programme involved free condoms and dispensed them at clinics. These consumers are outside the scope of the CSM programme, so I excluded these respondents. There were only 37 in this category. (c) The price was not applicable, because the individual did not purchase condoms. This was because: condoms were never used in sex; sex never took place; or because females do not usually purchase the condoms (the man does). The last reason might suggest the existence of sample selection bias. But note that the male ratio was not too much higher in my sample (77% of the full sample, while it was 86% in my sample). Males greatly dominated in both samples and one would want this to be the case in any evaluation of a CSM programme in Africa. Also, because I did not know who were zero purchasers, which could have been the reason why some did not respond to the price question, I could not do a Tobit estimation of zero and positive purchasers and hence estimate the intensive and extensive margins of the price subsidy programme. Instead, as explained in the text, I estimated a demand curve conditional on someone buying one pack of condoms and came up with a reason (i.e. quality differences) why people should be WTP different prices for the one pack they usually buy.

⁶Note, however, that as far as income was concerned, there existed a number of proxies in the survey (such as ownership of bicycles, cars and TVs, etc.). We tested a number of these income proxies and they were either not statistically significant or did not have as much explanatory power as the other nonprice demand determinants (education and age).

⁷We did find in the survey a question asking how many packs the individual usually bought (presumably, at the usual price), but not many persons responded (around a third of our sample of 1272). Responses were recorded into five categories. When we converted responses to a continuous variable, we tried this as an alternative dependent variable. However, price was not significant using this specification.

⁸Variables that were significant, but could not be included with the three variables ln P, age and education were: whether a person was single or married, male or female and had sex or not with a commercial sex worker in the last 3 months. Interestingly, religion (Catholic or not; Muslim or not) was not significant. Condom quality was a significant determinant of ln P itself and so was not included in the demand function with ln P; but condom quality was used later to decompose log price – see section ‘Valuing condom quality’.

⁹As explained earlier, those who valued condoms higher (were willing to pay a high price) were those who were older and had not completed high school. The average for the sample as a whole therefore had a lower age (26.35) and a greater proportion that went to secondary school (0.40).

¹⁰To illustrate the process, take the demand curve estimates given by Equation D2 in Table 1. The original estimate of the constant term was 2.0203. The coefficient attached to age was 0.3198. Multiplying this coefficient by 28.6667 (which is the level at which we are holding age constant) produces a value of 9.1687. Adding this value to 2.0203 results in an augmented α measure of 11.1893. The adjustment for the slope measures follows along similar lines. Consider Equation D5. The original slope (coefficient of log price) was −2.5217. The coefficient attached to log price times age was 0.0452. Multiplying this coefficient by 28.6667 produces a value of 1.2957. When we add this value to −2.5217 this leads to an augmented β of −1.2260.

¹¹This was the official exchange rate prevailing in December 1999–January 2000 as reported by Kihinga (2000) in the PSI Female Condom Consumer Profile Survey, p. iii.

¹²With constant costs, the marginal cost equals the average cost.

¹³It is interesting to note that, if we had taken the higher cost figure of 604 TZ shillings for optimal pricing, rather than the baseline cost of 290 TZ shillings, then outcomes would still have been strongly positive, with BCRs around 1.7 without the external benefits and around 2.6 with them.

¹⁴In the PSI sample, most females relied on their partners to buy the condoms for them. So gender was not just a simple demographic demand characteristic for it also reflected a particular mechanism of service supplied.

¹⁵Pindyck and Rubinfeld (1991) suggest that it is more efficient to use the actual rather than the predicted log price in Equation 2c. However, our conclusion remains the same with this specification as the relevant p-value here was 0.257.

¹⁶Not surprisingly then, 2SLS estimates of the seven demand equations (not reported) were very close to the OLS estimates.