Closed wleoncio closed 3 years ago
- Although it can be inferred that univariate/multivariate normal distributions are used with “c_mean”, “sigma”, and “cor_matrix”, it's better to include those information.
Addressed on 7b4ecff68e0ae42747faed3b196241b7d7b37871.
- When “c_mean” is not specified, the generation of mean is not given on whether they're freely generated or 0 by default.
- When “sigma” is not specified, the generation of sigma is not given clearly on whether it's randomly generated or 1 by default.
Addressed on 6c69a7b660dc8fdb36f5304e151bb9f738c9a8c5.
- Could not extract results from “summarize_cluster” function, e.g., extract certain statistics from certain level. It would be better if those could be extracted.
- Output from “summarize_cluster” function can't be saved as an object. It would make more sense if it can be saved.
- If (4) and (5) can be resolved, checking errors from simulation could be achieved via vast simulation instead of eye balling.
Addressed on 6f4c10c8690e068ffd268692abf71ffb17ec5847.
- When the means/sigmas are very small (e.g., 0.005) and with multiple levels, the estimation will be not accurate (e.g., school level estimates will be not as accurate as the student levels)
- Higher hierachical order => Worse estimation.
Addressed on 01536c162846181c2105c9abf53518846deae251.
- The output for “sigma” only includes two digits, so it’s impossible to compare the results if the input values are less than 0.01
@Hugo-v587, here's a counter-example:
cluster_gen(c(1, 10), n_X = 1, n_W=0, full_output=TRUE, sigma=pi)
Notice how all standard deviations are 3.141593
as specified, so the output uses as many significant digits as R is configured to print by default. Can you confirm this works on your end?
- Changing the seed will generate very different estimation results.
@Hugo-v587, I wonder if this could be a result of a small sample size. Please report if increasing the sample size doesn't fix this (with accompanying code).
All reported errors were addressed on f44d28dd589a6f1129040192a775b6db26d87702.
0. Setup
I've tested most values below. Not all testings are shown in this report. I only included the testings that are showing errors/warnings or inconsistent results.
3. c_mean, sigma
Overall suggestions
Error and warning messages
This warning message is due to the lack of sample size. It showed error but could still run.
initial correlation inadmissible, -1.00337997759108, set to -0.9999could not compute polyserial correlation between variables 2 and 1 Message: Error in optim(rho, f, control = control, hessian = TRUE, method = "BFGS") : initial value in 'vmmin' is not finite
[x] The third version of error message regarding the non-finitite value: NaNs producedNaNs producedNaNs producedNaNs producedcould not compute polychoric correlation between variables 10 and 8 Message: Error in optim(0, f, control = control, hessian = TRUE, method = "BFGS") : non-finite finite-difference value [1]
[x] Warning messages after "Heterogeneous correlation matrix" will make the "q1 q2 q3..." misaligned with the correlation matrix, check {r mean}