ANN ARBOR—Researchers have known for decades that a positivity bias exists in language where people tend to use more positive words than negative ones. What they haven’t been able to pinpoint is whether this phenomenon is affected by time and events.

University of Michigan researchers Rumen Iliev and Robert Axelrod, along with Joe Hoover and Morteza Dehghani of the University of Southern California, used automated text analysis tools to find out if positivity bias is a static property of language or if it varies as a function of time.

“We found that positivity bias in American English was not static, but has changed over time,” said Iliev, who worked on the analysis while a postdoctoral researcher at U-M’s Ford School of Public Policy. “We found that in objectively worse environments, such as during wartime or economic misery, language positivity declined.”

The results are published in PNAS, the Proceedings of the National Academy of Sciences.

Axelrod, U-M professor of public policy and political science, said the research revealed that changes in language positivity were predicted by changes in psychological states, such as happiness.

“We also found a general decrease in language positivity, which was consistent with reports of decreasing pro-sociality in American society,” Axelrod said. “The results we observed suggest that positivity bias is related to multiple factors including objective circumstances, subjective states and societal changes.”

Iliev explains how positive bias can be a problem, for example, in restaurant reviews. Suppose the only data you have is a large text file with thousands of reviews written by the customers of this restaurant. Since reading reviews individually is very costly, you might want to write a small computer program which searches for affective words, and then you can compare the counts of all positive words (e.g., fantastic, great, amazing) to the counts of all negative words (e.g., horrible, tasteless, disgusting).

“The problem with this intuitive solution is that you need to know the general prevalence of positive versus negative words,” he said. “It turns out that, that, on average, there are more positive than negative words both in speech and in writing, so a greater prevalence of positive words in our reviews might simply be a linguistic artifact rather than a genuine positive sentiment.”

Research increasingly shows that psychological states are reflected in both positive and negative language, said Morteza Dehghani, USC assistant professor of psychology and computer science.

“Even though the prevailing view is that the aggregate happiness trend in the United States is static, our results provide strong indirect evidence that this trend is actually downwards,” Dehghani said. “This may have important policy implications regarding the overall well-being of our society.”

Dehghani said that scientists should study positive and negative language in other nations with different levels of happiness to see whether the U.S. findings are universal or unique.

“For example, it would be interesting to look at Denmark or Norway—places that tend to score high on overall happiness, and see whether we find the same negative trend,” Dehghani said.

More information: 

• Robert Axelrod