Language of Inflation Targets

In an online survey, we will explore how people interpret different language used by the Federal Reserve in reference to its inflation targeting goals and we will explore how this different language impacts individuals' expectations about the path of inflation.

In an online survey, we will explore how people interpret different language used by the Federal Reserve in reference to its inflation targeting goals. In particular, the Fed uses different language when it refers to its inflation goals, and some of this language may engender different interpretations than what the Fed intends. To study this, we will provide subjects with different versions of inflation target language and solicit their interpretations of what this language means in terms of what the Feds goals are. We will also use an incentivized question to explore if this different language impacts individuals' expectations about the path of inflation.

2024-09-30

2024-11-30

We are interested in the following primary outcomes:
-subjective scoring on how closely each interpretation provided aligns with their interpretation of different Fed language
-the proportion of subjects responding that one version of inflation target language will engender a faster decline in inflation that a different version of inflation target language.

This is not relevant for our study as we will be reporting the main variables directly.

As secondary outcomes we may also explore how responses differ across demographic groups, and subjects' economic knowledge (as measured by correct answers to some inflation questions)

This is more of a descriptive project and less of an "experiment" as there are no "treatments" per say. We are simply interested in how individuals interpret different language and if different forms of language impact expectations.

Since there are no treatments in this survey randomization is not very relevant. We will randomize the order that subjects see different answers/interpretations to control for any order effects, and this randomization will be done by Qualtrics.

The only randomization is the ordering of answers/interpretations and this will be done at the subject-level.

No

300 (individuals will respond to multiple questions)

300 (individuals will respond to multiple questions)

No treatment arms, so sample size is 300

The statistical test we care most about is if the different versions of language will impact the proportion of subjects expecting subjects to expect inflation to decline faster in one context vs. another. In this multiple choice question there are three options, one is a neutral result, and the other two are either than one version of language (call this language A) will result in a faster decline in inflation (relative to language B), and the other option is that language A will result in a slower decline in inflation (relative to language B). To see if the language matters, we are thus most interested if one of these options is selected more frequently than the other, indicating subjects perceive one version of the language to result in more rapid declines in inflation. A power calculation from a Chi-Squared test using probabilities from these answers from a pilot (N=30) gives statistical power of 1 with N=300. So we are confident N=300 is sufficient power-wise. The other statistical tests will be comparing the interpretations scores across the prompts. You could think of the different prompts as different "treatments" (all within subject variation). To see if the different prompts engender different interpretations we can conduct a paired t-test. From the pilot data power calculations suggest we would need small sample sizes to achieve power of 0.8 or greater. For example, we are interested in how the interpretations scores for interpretation 3 are different conditional on the prompts. From the pilot data, comparing scores for interpretation 3 across the different prompts yields the following sample sizes for alpha=0.05 and statistical power=300: N=3, N=4, N=3, and N=5. With a Bonferroni correction for, say, these 4 tests, gives adjusted sample sizes of N=4, N=5, N=4, and N=7. So we are extremely well powered with N=300.