I want to create indices for several dimensions ( based on many indicators) and then a total index averaging dimensions using STATA. My data is panel data; several countries for several years. so each country will have an index for each year.
1- is there a specific method for dealing with panel data in stata or is it similar to the conventional method? do I need to identify the data as a panel in STATA? ( will the output be the pc for each country each year?)
2- is it right that I shall run PCA for each dimension separately?
2- does stata standardize the data when performing PCA or do I need a specific function?
4- if I want the index to be on a scale (0-100). is there an option for that or I need to define a new variable with a min/max equation?
Dear Tara Vanlı
1. As far as I know you need to carry out such an operation spefically, for instance, using a loop. The details depends on what you want (e.g., you might want that each year and dimension is on 0-100 scale or that the dimension is 0-100 considering all years in the same country, etc.). It depends on what you want... You can customize easily what you want using "if" and loops "foreach" and "forvalues".
2. Again, it depends on what you want.
3. Stata pca function uses the correlation matrix and data are standardized.
4. You need to normalize yoursefl using the min-max transformation. You can do it yourself using loops and perhaps temporary variables (https://www.statalist.org/forums/forum/general-stata-discussion/general/1314401-normalising-a-variable-to-be-between-0-and-1) or try NORM command (https://ideas.repec.org/c/boc/bocode/s457698.html).
I hope it helps,
José-Ignacio
Dear Tara Vanlı,
PCA is just a data reduction technique when you want to work with a smaller number of dimensions, creating new variables (components). Such components are determined according to the variability and correlation observed in the original variables (dimensions). So it is very important that your variables to be harmonized. But the way you put it, I wonder if it would be better for you to work with SEM or GSEM, where you can calculate factors by dimension, year and "secondary" factors based on the factors of each dimension.
Dear Tara,
Answer to Qn #1.
The method for dealing with panel data in Stata is similar to the conventional method. Even if it is panel data, Stata automatically recognises that the data is panel in nature hence you basically do not need to identify or set the dataset to panel format. (If you do not restrict your observations within a particular range, the PCA output will be for all countries and all years. Thus, if you want PCA output for each country and corresponding time span, you have to restrict your observations within the relevant range).
Answer to Qn #2.
PCA is a data reduction technique, hence running a PCA for each dimension is advisable in exceptional scenarios such as when the respective dimension contains a reasonable number of items based on expert judgment.
Answer to Qn #3.
Stata, by default, standardizes the data of all variables when performing PCA and creating loadings.
Answer to Qn #4.
PCA assumes that all underlying variables follow a normal distribution.
To provide an informed suggestion/answer to your Qn #4, what is the nature and measurement level of your data?
All the best!
DearTara
Number of indicators and number of observations are not specified clearly. The purpose of developing indicators is also not mentioned. PCA can also be used for the construction of index.Index can be constructed for each dimension separately after standardization and weighted average index depending on the number of indicators in each dimension may more meaningful than simple average index. Since the data is panel, 1) the index construction of indices for countries is recommended with same indicators for different dimensions
2) the average of all the indicators in each dimension of each country and standardize within country initially.
3. Perform PCA on standardized values of the indicators using STATA ( don't use the correlation matrix) for a year and repeat the same for different years. Use the loading's of of indicators for the construction of Index in each dimension.
4. DO loop in STAT can be used for the construction of indices for several dimensions and years
Dear Tara,
Yes. Panel data is for several years and several countries.
Q1: You can employ PCA and note that STATA will automatically recognize your inputs as panel data.
Q2: Yes. You should run PCA for each dimension as it is specific for a reduction technique.
Q3: Yes. STATA standardizes your variables.
Q4: Generally, PCA asumes a normal distribution.
Wishing you the best.
Regards
Dear Tara Vanli,
An additional care. PCA is a data reduction technique for continuous indicators. For categorical indicators you need to use the correlations of the polychoric or tetrachoric commands. Other alternatives are MCA (Multiple Correspondence Analysis that can be thought, to some extent, as an extension of PCA for categorical data) or GSEM with categorical variables in the measurement equation. As STATA makes it very easy to use PCA, MCA and GSEM, so you need to work the inputs well to get the appropriate response, paying attention to the types of variables and harmonizations between them etc.
Leonardo Oliveira Kehinde Mary Bello Brigit Joseph Elvis Munyaradzi Ganyaupfu José-Ignacio Antón thank you very much for taking the time to answer. your valuable answers were very helpful
- Badges
- Science topic
- Similar topics
- Filing