Dear all,
I have the following data and questions related to the issues in the results.
Independent variable
I did an online survey and asked parents questions about neighbourhood-built environment factors like neighbourhood type, accessibility, and neighbourhood safety. These are major factors that have sub-questions inside them. These questions are on a five-point Likert scale. Therefore, the parents’ responses are on a scale of 1 to 5. These are my independent variables.
I have calculated the average score based on the response to each question by the parents under the major factor. For example, in the neighbourhood type factor, I asked five questions. Each question was on a five-point Likert scale. So, If Q1 = 5 (response given by parent), Q2 = 3, Q3 = 2, Q4 = 1, Q5 = 1.
Then the subscale score for neighbourhood type was = 5+3+2+1+1 / 5 = 2.4
A similar procedure was adopted for other factors. The final score for each factor was taken as the binary categorical variable, with a cut-off value = 2.5.
Dependent variable
I also asked questions about the physical activity of their eldest adolescent kid. Adolescents' response was asked in the number of days in a week. This is my dependent variable, and I have taken it as a continuous variable.
Total data
I have collected this data for three cities in India. So, for the final dataset of the model, I have combined all the data points into a single Excel sheet.
Aim
I aim to understand the effect on adolescent physical activity due to the parents’ physical activity and built-environment factors. Therefore, I have used a multilevel linear regression model to investigate the relationship between children's PA and neighborhood-built environment. In this model, I have taken the city variable as a random effect variable and the adolescent's age & neighbourhood residency duration as a fixed effect.
Results
The results are unexpected, and I am not convinced. They are also statistically insignificant.
My questions are as follows:
3. For this type of data, Is it correct to use a multi-level linear regression model for this type of data? As all the independent variables are fixed.
I would be grateful for the discussion and input to help me in this analysis.
warm regards
Laxman
Laxman Singh Bisht, your approach of averaging Likert scale data and converting factors into binary categories has some limitations, particularly regarding potential loss of information and arbitrary cut-off values. While a multilevel linear regression model is suitable for analyzing data from multiple cities, it may not be necessary if your independent variables are all fixed. Consider whether using ordinal regression techniques for Likert scale data and a standard linear regression model might be more appropriate. To discuss with me in person, let's connect on WA; https://wa.me/+923440907874.
Before testing the hypothesis, it is mandatory to test the scale measuring instrument, namely the validity of each item and the totality of items, namely reliability with Cronbach's alpha.
Hi,
These are my suggestions:
1) Taking the average of the Likert items is OK.
2.1) Recoding the Likert average scores to binary variables is unnecessary. If you have good reasons to do that, consider the cut-off value in the middle of the scale. From 1 to 5, the middle point is 3, not 2.5.
2.2) You cannot use multilevel "linear" regression after recoding to binary variables. Now, you have to use multilevel "logistic" regression.
3.1) If you want to generalize to the population of cities (second-level units), using a multilevel linear regression model is OK. But, using this approach, you can consider testing for random effects for your level-1 predictors.
3.2) If you only want to control for the cluster effects associated with cities, consider using linear regression (not multilevel) and adjusting the standard errors by viewing the cities' cluster effects.
Best,
Roberto
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India