Hawthorne effect treatment in LENA Home data

The Hawthorne effect is the human tendency to alter behavior - for example, to overperform - when we know we are being observed. In the context of a LENA Day, the tendency would be for the adult to talk/interact more than normal, thereby creating skewed results that are difficult to replicate or sustain, much less to improve upon.

To counteract the Hawthorne effect in LENA recordings, we automatically reduce the Words and Turns counts on the first valid, non-hidden recording. Unless a program has custom hide rules implemented by LENA staff, or an admin user is manually hiding recordings according to their own criteria, this would typically be equivalent to the first recording that is at least 9.5 hours in duration.

  • The daily Adult Word Count (the 12-hour projected value) is reduced by 15%.
  • The daily Conversational Turns Count is reduced by 10%.

We chose these numbers based on an analysis of our normative data. Comparing the first recording to the remaining recordings within family, we saw an 8% elevation in Adult Word Counts for the first recording, and a 10% elevation in Turns, both of which were statistically significant. We know that there are two things that could influence parents to talk more than they typically do on their first recording in LENA interventions: 1) novelty of the recorder, 2) knowing their reports will be seen by a home visitor or coordinator and wanting to perform well due to social pressure. Normative participants would have been influenced only by the first issue (novelty) because they never saw reports.This being the case, we increased the adjustment for Words from 8% to 15%. Once we had settled on 10% adjustment for Turns and 15% for Words, we looked back at our normative data to see how many of them show this kind of elevation on the first recording. Our results indicated that 44% of our normative sample had 15% elevations on words, and 45% showed 10% elevation on turns, so we went with these values. We have examined the observer effects over time in different settings and will continue to do so and will factor that into how the adjustment is implemented and how we factor observer effects, if at all, in offline data analyses.

How can I identify the first valid non-hidden recording for an individual?

To identify the files in bulk for analysis, use the full-day recordings export, which is automatically sorted by participant and recording date (ascending). For each child, locate the earliest recording date where both HiddenRec and FilteredRec = FALSE.

For a more casual approach to the individual child, you can also look at the report to find the left-most colored bar.

How can I see what the original values were?

It's fairly simple to calculate both the actual and projected counts for a child’s first valid non-hidden recording using the data exports.:

To find the original projected counts, use the full recording export. Divide the reported AWC value by 0.85, and the reported CTC value by 0.90.

To find the original actual (raw) counts, sum the hourly counts from the hourly export. Note that the resulting value is not adjusted for duration and so is not appropriate for comparison against the 12-hour norms, nor against other recordings of varying durations from the same child or other children.