Employment Is the Harder Outcome

The Delhi experiment's final test is whether changed beliefs and search routes translated into work. This chapter reads the twelve-month evidence on employment, work intensity, earnings, and occupation, then asks how much confidence the robustness checks deserve. The result is narrower than a success-or-failure verdict: overall employment did not significantly rise, but personalized information was associated with later gains in earnings and salaried work for some women. Understanding that difference is essential for judging what the intervention changed and what it left untouched.

Employment is a demanding endpoint because it sits after the stages already observed. A person must notice information, register or find another route, search, apply, receive a response, and obtain work. Even then, the quality of that work can vary. A programme can change the probability of being employed, the number of hours worked, the kind of occupation held, or the earnings received without moving all of those measures together.

Employment did not move on the broad margin

The study follows labor-market outcomes monthly from the intervention month through twelve months afterward. The extensive margin asks whether a person is currently employed at all. Across the full sample, pooled treatment effects on current employment were statistically insignificant throughout the period. Neither the general-message nor the personalized-message group produced a statistically significant overall employment effect. The point estimates for Personalized women turned positive in later months, but the broad comparison did not establish a significant rise in employment for the sample as a whole.

“Statistically insignificant” needs a precise reading. It means the study did not detect a sufficiently clear average difference from the control comparison at the chosen level of uncertainty. It is not proof that the true effect is exactly zero, nor proof that information can never create employment. The estimate may be too small, too variable, or too dependent on subgroup and timing for the sample to distinguish it confidently from zero. Conversely, a positive point estimate alone is not evidence of a reliable effect.

The same caution applies to work intensity. The researchers examined hours worked per day and days worked per month, but treatment effects for the full sample were insignificant on both margins. Some gender differences appeared in the first two months, including lower work hours for men relative to women in the Non-P group and higher hours for men in the Personalized group relative to the Non-P group. Those early contrasts were not a general increase in work intensity. They show that a labor-market intervention can produce short-run differences in how people work even when the aggregate employment measure remains flat.

The absence of an employment increase is economically plausible. The intervention primarily changed information available to job seekers and the routes through which they searched. It did not directly create vacancies, alter every employer's screening process, or guarantee that a relevant job would be available at the desired wage and location. If the participant-side information problem is only one bottleneck, removing it can change search without moving the final count of people working.

Earnings and occupation tell a narrower story

The more revealing results appear on the intensive and occupational margins.