WEBVTT

00:00:01.800 --> 00:00:03.520
Every year, the HILDA Survey asks

00:00:03.560 --> 00:00:06.060
thousands of Australians about their lives.

00:00:06.100 --> 00:00:08.290
Some questions, they answer alone.

00:00:08.330 --> 00:00:09.970
One is about drugs.

00:00:10.570 --> 00:00:12.490
For decades, economists have asked

00:00:12.530 --> 00:00:15.290
whether drug use costs people at work.

00:00:15.330 --> 00:00:18.670
But people who use drugs differ in other ways.

00:00:18.710 --> 00:00:22.330
Some don't say. And only workers have wages.

00:00:22.630 --> 00:00:24.990
Every household also meets an interviewer.

00:00:25.030 --> 00:00:26.930
They don't ask about drugs.

00:00:26.970 --> 00:00:28.110
Yet what people report

00:00:28.150 --> 00:00:30.690
shifts with who the interviewer was.

00:00:30.730 --> 00:00:32.710
Survey researchers call that error.

00:00:32.910 --> 00:00:34.630
This study uses it.

00:00:35.330 --> 00:00:36.870
Within an area and year,

00:00:36.910 --> 00:00:38.210
who gets which interviewer

00:00:38.250 --> 00:00:40.530
comes down to fieldwork rosters.

00:00:40.570 --> 00:00:42.030
If that's as good as random,

00:00:42.070 --> 00:00:46.450
and interviewers affect pay only through reporting,

00:00:46.490 --> 00:00:48.390
the trace becomes a lever:

00:00:48.430 --> 00:00:51.750
a second way to ask the same question.

00:00:52.150 --> 00:00:53.710
For cannabis, both ways agree:

00:00:53.750 --> 00:00:57.070
about four percent lower hourly pay.

00:00:57.170 --> 00:00:59.930
For frequent methamphetamine use, they split:

00:00:59.970 --> 00:01:03.070
slightly higher pay, if anything, one way;

00:01:03.110 --> 00:01:05.430
almost ten percent lower the other.

00:01:05.470 --> 00:01:07.690
Corrected for testing six drugs,

00:01:07.730 --> 00:01:10.210
it narrowly misses the usual bar:

00:01:10.670 --> 00:01:12.850
a lead, not a verdict.

00:01:12.950 --> 00:01:14.930
For frequent inhalant use,

00:01:14.970 --> 00:01:17.430
the estimated penalty roughly doubles.

00:01:17.730 --> 00:01:20.730
Six drugs. No single answer.

00:01:20.830 --> 00:01:22.970
Some findings hold either way.

00:01:23.010 --> 00:01:26.110
Others depend on where the evidence comes from.

00:01:26.410 --> 00:01:28.330
The noise was never just noise.