# Create a new scale from the averages of 7 items

As a researcher on a study, hypothetically, I want to create a scale assessing how much nurses who were providing double duty care giving were blurring boundaries between their duties at work and their care giving at home. They included 7 questions, all in the same ordinal (likert) type measurement of strongly agree, agree, neutral, disagree and strongly disagree, related to this extent of additional care giving at home.

I must create a new scale for the researcher by taking the average of each of those 7 items and then providing some descriptive statistics for the new scale, called the Care Giving Interface, which is meant to assess the overall extent of double-duty care giving that nurses experience. As part of my scale analysis, determine how well the 7 items appear to measure the same overall construct of the extent of double-duty care giving.

How can I take my 7 items and combine them into one new variable? I know that once I make the new scale, I can do a descriptive statistical analysis and generate the overall mean etc, but I do not know how to combine the various items into one variable.

• Are you saying that you have measurements of 7 variables (responses to questions) and you want to "merge these 7" to form one index? What is then the purpose of this index, is there some (single) outcome variable that the index should predict? – Superpronker Dec 13 '16 at 19:21
• "I do not know how to combine the various items into one variable" — But you just said "I must create a new scale… by taking the average". So what's your question? – Kodiologist Dec 13 '16 at 20:08
• So, I want to create a new scale assessing how much nurses who are providing double duty care giving blur the boundaries between their duties at work and their caregiving at home. So, yes, I have 7 questions (items) all related to this extent of caregiving at home. I must create a new scale for the researcher by taking the average of each of those 7 items and then provide some descriptive statistics for that new scale which I will call the 'Care Giving Interface.' Which is meant to express the overall extent of double duty caregiving that nurses were experiencing. – Erika Dec 14 '16 at 14:05
• Would it be appropriate to use a Linear Regression to assess my Likert scale outcomes? – Erika Dec 14 '16 at 14:11

Below outlines an approach (see also: Combining several Likert items from a Likert scale into one variable)

+----+----+----+----+----+----+----+----+-----+
|    | Q1 | Q2 | Q3 | Q4 | Q5 | Q6 | Q7 | CGI |
+----+----+----+----+----+----+----+----+-----+
| P1 |  1 |  4 |  3 |  3 |  4 |  2 |  0 |  .  |
+----+----+----+----+----+----+----+----+-----+
| P2 |  . |  . |  . |  . |  . |  . |  . |  .  |
+----+----+----+----+----+----+----+----+-----+
| PN |  . |  . |  . |  . |  . |  . |  . |  .  |
+----+----+----+----+----+----+----+----+-----+


Map the responses to a numeric range, e.g. [0,4]. Organize your data, for example, as in the table above ($Q_1$ - $Q7$ are the questions, $P_1$ - $P_N$ are participant responses). To get an expected value for each question, $\bar{Q}_i$, average over the rows. The root mean square deviation will give you a single value for each participant - call it Care Giving Interface ($CGI$).

$$CGI_N = \sqrt{\frac{\sum_{i=1}^7(\bar{Q}_i - Q_i)^2}{7}}$$