With 28 levels you are implying that different studies use different WTP categories and you are pooling across studies. I am fairly certain that simple pooling is impossible. Suppose that one study used intervals [50, 100) and [100, 150) and another used [50, 75), [75, 100), [100, 125), [125, 150). You can consider all of the lower/upper value pairs as representing interval-censored values, and get the Turnbull estimate of the cumulative distribution function. This will not have steps at all possible left and right endpoints. If you have covariates this generalizes to a proportional odds model with interval censoring as handled by R rmsb::blrm (Bayesian using Stan) and rms::orm (frequentist).
Whichever approach you take, without having uniformity of intervals there is not reliable way to do this simply.