In my research, several variables consist of 2 items. For example, one is named as "screen time". the items are:
1) "On a normal weekday during term time, how many hours do you spend watching television programmes or films?"
2) "On a normal weekday during term time, how many hours do you spend playing electronic games on a computer or games systems, such as Wii, Nintendo D-S, X-Box or PlayStation?"
I have data from 4 different samples (cohort study) and the inter-item correlation scores are ranged from .18 to .28. The Cronbach alpha scores are lower than .5.
Do you think it is enough to use the inter-item correlation for the reliability or what else can I do?
If the inter-item correlation is .18, one item explains 0.18^2= 3% of the variability of the other. I would argue that is far too low for claiming they are one scale, and the two items are effectively measuring two rather different things.
It depends on the nature of the scale. If the items are meant to reflect (ie caused by) some underlying construct, then higher correlation between the two items is desirable, so that the Cronbach alpha (or also split half, in this case) reliability estimate is .7 or higher. However, there are cases where the sum of the two items defines a meaningful construct even construct, even if the correlation between the items is negligible, or even negative.
For example, each item obtains information on the amount of time spent daily on different types of exercise, the sum would be the total time each day spent exercising, which is a meaningful construct. However, there, could be a trade-off between the times spent on each activity, which would tend to lower the correlation between items, or even become negative.
In structural equation modelling, these later types of scales would correspond to what are called "formative" latent factors, in contrast to "reflective" factors. Higher correlations between markers are only desirable for ones defining reflective factors.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Probability
- Probability Theory
- Reliability