MISLABELING ESTIMATES
We use statistical models to estimate mislabeling rates.
Compared to taking the average of the number of times a product has been mislabeled (# mislabeled samples divided by total number of samples tested), our approach has important advantages for characterizing seafood mislabeling.
Bayesian meta-analysis
Bayesian meta-analytic approaches offer a number of advantages to characterizing seafood mislabeling, including delivering easily interpretable results with fewer assumptions.
EFFORT
First, it takes effort into account. This is important, especially when samples sizes are small. For example, is it not uncommon to flip a coin four times and get three tails. But, that does not mean the probability of getting a tail is 75%. Using the same analogy, if a study tests four Pacific Salmon samples for mislabeling and three turn out to be Atlantic Salmon—that does not mean the mislabeling rate for Pacific Salmon is 75%. If we flip a coin 400 times, we know that the probability of getting tails converges to 50%. Our statistical models act in a similar way, giving more weight to studies with more samples.
UNCERTAINITY
Second, our statistical models not only estimate mislabeling rates—they also estimate the uncertainty of those rates. That is, the precision of the estimate. Sticking with the coin toss analogy: precision is the degree to which you get the same results if you repeated a four (or 400) coin flip experiment multiple times under the same conditions. Not surprisingly, you are much more likely to get the same result repeating 400 coin flips compared to four coin flips.
Fore each product, we provide a global mislabeling rate and its uncertainty. In this example, the global mislabeling rate is somewhere between <1% and 12%, with the most likely value being 2%.
For each product, we provide the most likely mislabeling rate given the current data. We also provide a minimum and maximum likely value. These values can be interpreted with the following analogy: if you re-tested the same number of samples under the same conditions, there is 95% chance the mislabeling estimate would be between the minimum and maximum values. Thus, the greater the spread between the two values, the more uncertain the mislabeling rate.
We also provide global production estimates for seafood products from the Food and Agriculture Organization of the United Nations. Mislabeling rates must be coupled with other data, like production, in order to understand potential ecological and economic consequences.
Production
Production varies by product and it represents its availability in the market.
The data, methods, and results presented here have been subject to peer-review and published in the scientific journal Biological Conservation.