Experimental Design

To summarize the interventions listed above, we have a total of 9 experimental groups.

• The control group (“G0”), which is constructed to best resemble the conditions of annuity choice that people typically face. These annuities are worse than actuarially fair, the status quo is not owning an annuity, respondents are not induced to think about savings nor are the reminded of their savings plans for each contingency, and the annuity choice uses natural wording such as “annuities”, “insurance” or “Social Security.”

Five treatments investigate the three mechanisms of primary interest:

• One treatment group which is like the control group but where the status quo of not owning an annuity is removed (“G1”)

• Four treatments that increase proper accounting for contingent savings decisions, leading to the following groups:

o Savings salient group (“G2”)

o Explicit contingencies group (“G3”)

o Explicit contingencies, No context, group (“G4”)

o Explicit contingencies, No context, Dominance, group (“G5”)

• The reverse correlation treatment group (“G10”)

Two treatment groups to help us gauge the magnitude of the response:

• The low-price group (“G20”)

• A combined treatment group (“G35”), which combines a low price, the treatment in G5, and reverses the correlation as in G10.

We randomize subjects such that the expected number of observations in each of these 9 groups is expected to be equal.

Each respondent makes 6 decisions, and the decisions are divided into three blocks. The order of the blocks varies across respondents.

Savings block (3 decisions):

• The desired amount of savings if the respondent does not have an annuity

• The desired amount of savings if the respondent has a better-than-fair annuity (low price)

• The desired amount of savings if the respondent has a worse-than-fair annuity (high price)

Regular annuity decisions (2 decisions):

• Annuity choice if the annuity has a low price (i.e., is better than fair)

• Annuity choice if the annuity has a high price (i.e., is worse than fair)

One of the three “Explicit contingencies” decisions:

• Annuity choice if the contingencies are made for the treatment (i) explicit contingencies, (ii) explicit contingencies, no context, or (iii) explicit contingencies, no context, dominance. Because contingencies can only be made explicit if the respondent has already made their savings choices, this block necessarily comes after the savings block.

The primary randomization allocates annuity decisions (not respondents, because respondents each make three annuity decisions) into one of the 9 experimental groups described above. Decisions are randomized into these 9 groups with equal probability.

All decisions of a given respondent are either randomized into “regular correlation” groups (in columns A or C of Figure 1 in the uploaded "Full summary of experimental design" ) or into “reverse correlation” groups (in columns B or D of Figure 1 of Figure 1 in the uploaded experimental design summary).

There are five secondary randomizations:

• Wording. In the explanation of the experiment and in the some of the annuity decisions (the ones with context), the annuity choice is described either in terms of “annuities,” “insurance,” or “Social Security.” One of these three wordings is selected with probability 1/3 for each of the respondents whose annuity decisions are randomized into “regular correlation” groups. Only the insurance wording is used for respondents whose annuity decisions are randomized into “reverse correlation” groups (the other two wordings would be unnatural).

• When there is no status quo, we randomize (with equal probability) which option is presented on the left or the right.

• If the savings block is asked first, the order of the two annuity blocks is randomized.

• The order of the 3 savings decisions in the savings block is randomized

• The order of the two annuity decisions in the regular annuity decision block is randomized