Item Parcelling for Measurement Models

Item Parcelling = Combine the items in a factor and use them as indicators

Parcelling is a compromise between the original SEM (with all items intact) and a structural model with composite scores (where respective items are totalled or averaged to arrive at a factor’s score).

Three ways to parcel:

1)    Assign the items in a factor randomly into 3 or 4 parcels

a.    A good option if all items have high loading

2)    Distribute items with high loading across 3 or 4 parcels

a.    A good option if some items have low loading

3)    Use a theory to create dimensions within the factor

In this example, we measure Self-Compassion as a latent variable using 9 indicators (Item1 to Item9)

Figure 1. Original measurement model with 9 indicators

We can do item parcelling by assigning the items randomly to three parcels. In this example, the first parcel is made of Item 1, 4 and 5, the second parcel is made of Item 2, 3, and 9 and the rest of the items made the third parcel. You could use a random number generator to get the numbers 1 to 9 as select the first three, second three, and the last three numbers.  Notice for this example, we simple sum up the three items within each parcel. Once we have these parcels, we can calculate the estimates in AMOS as usual.

If we want to parcel using the second method, we can list the items based on their loading (from biggest to smallest). Then, three items with the highest loading will be assign to each parcel. The three items with the lowest loading will be assigned to each parcel. The three items with the middle loading values will accompany the top three and bottom three in each of the parcels. Using the same diagram below, we can assume that Items 1,2 and 6 has the highest loading, while Items 5, 9 and 8 has the lowest loading.

Figure 2. A measurement model with 3 parcels comprising summed items

In the second example (Figure 3), the parcels are made of averaged score of the items.

Figure 3. A measurement model with 3 parcels comprising averaged items

Harris Shah Abd Hamid

18/12/2019