How reliable is LENA? : validating and inspecting Language Environment Analysis in German

In recent years, the impact of early audio environment of infants and children has gained increasing attention in research. This leads to a need for cost-efficient solutions to analyze these audio environments. LENA – Language Environment Analysis – is a device and software aiming to do exactly that and is commonly used in research and clinical application. However, its reliability has not yet been tested in German, which is a research gap this study aims to fill. Manually coding 144 5-minute audio samples of 24 Austrian children showed Interclass Correlation Coefficients (2,1) of 0.81 for Adult Words, 0.76 for Child Vocalizations and 0.55 for conversational turns. This falls in line with the results of other validation studies, that also showed good results for Adult Words and Child Vocalizations, but only moderate results for Conversational Turns. Additionally, dividing the sample into three different, equally sized age groups (9-12 months, 18-21 months and 27-30 months) and statistically comparing ICCs revealed significant differences in Child Vocalization ICCs between the 9-12 months group and the 18-21 months group (p = 0.0010), with the younger group showing only poor correlation between human and LENA counts. Furthermore, a qualitative analysis of audio files where severe differences between LENA and human counts were noticeable, highlighting several other issues, such as LENA’s performance being worsened by day-to-day activities like driving in a car or the recorded child running and playing. Despite this, the overall results are satisfactory for use of LENA in German speaking areas, as long as the moderate result of Conversational Turn counts is kept in mind.