Park-Levine Probability Model

The Park-Levine Probability Model is a formula showing how deception detection accuracy varies as a linear function of the truth-lie base rate. The truth-lie base rate has to do with the percentages of truths and lies being judged in a deception detection task, and accuracy is the percent of correct truth-lie discriminations in a deception detection task. Truth bias is also an important consideration. Truth bias refers to the tendency to judge messages as honest regardless of their actual honesty. So long as people are truth-biased, the model predicts that as the messages under scrutiny become proportionally more honest, the accuracy of the judgments increases. Conversely, the model shows that the greater the proportion of lies, the lower the resulting accuracy. If the extent of truth bias and the accuracy for truths and lies are known for one specific base rate, for example 50 percent truths and 50 percent lies, the model can be used to determine what the accuracy would be at a different base rate, for example with 75 percent lies and 25 percent truths.

Specifically, the model predicts that the percentage of accuracy with which truths and lies are correctly discriminated will equal the accuracy for only truths, times the proportion of messages that are truthful, plus the accuracy for lies, times the proportion of messages that are lies. That is, accuracy = (truth accuracy)(truth proportion) + (lie accuracy)(lie proportion).

As an example, imagine that the overall observed accuracy in a deception detection experiment, in which judges were shown 10 truths and 10 lies, was 55 percent correct. Imagine further that the judges were truth biased and more accurate for truths than lies, getting 75 percent of the truths correct but only 35 percent of the lies right. Because there are an equal number of truths and lies, the truth-lie base rate and the proportion of truths and lies are all 0.50. So, there is an accuracy of 0.55 = (0.75)(0.50) + (0.35)(0.50).

If the base rate was 90 percent truths and 10 percent lies, the model predicts that accuracy would increase to (0.75)(0.90)+(0.35)(0.10)= 0.71, or 71 percent. In contrast, if the base rate were 90 percent lies and 10 percent truths, the model holds that accuracy would fall to (0.75) (0.10)+(0.35)(0.90)=0.39, or 39 percent accuracy. If 0.71, 0.55, and 0.39 were plotted, each accuracy figure would fall on a straight line.

The model predicts a linear relationship between the proportion of messages that are [Page 750]honest and deception detection accuracy. The slope of the linear relationship is a function of truth bias. The slope becomes steeper as truth-bias increases. Consequently, the truth-lie base rate makes a bigger difference when truth biases are strong. If judges were lie-biased, the slope would be negative.

Research has strongly supported the utility and predictive power of the Park-Levine model. In a paper published in 1999, Levine and his colleagues showed that people are generally truth-biased, and that accuracy increased linearly as the base rate increased from 75 percent lies to 75 percent truths.

In another study published in 2006, Levine and his colleagues offered a more precise test of the model. Truth and lie accuracies were measured in a control group with 50 percent truths and 50 percent lies. Then an experiment exposed judges to deception detection tasks with a variety of several different base rates ranging from all lies to all truths. Accuracies were observed in the experimental groups and compared to accuracies predicted by the model, using the base rate and the accuracies from the control group. The experiment showed that the Park-Levine-model predictions closely approximated the experimental results. The predicted linear relationship was observed, and all accuracies were predicted to within 2 percentage points. The model has also been supported by several experiments that are as yet unpublished.

...