I'm working on a large weighted national dataset and keep running into the following error when using the svychisq() function. The overall dataset has 7552 observations with 784 variables, however I am getting this error code for some subsets I created with 1,000-2,000 observations. The following code includes "NM.svy" as a weight data frame for one of those subsets.
svychisq(~RaceEth + BFDur, NM.svy, statistic = "Wald")
Error in solve.default(V[use, use], Y[use]) : system is computationally singular: reciprocal condition number = 4.34304e-22
I only run svychisq() to get a p-value score so would it be okay to use the reciprocal condition number of "4.34304e-22" to represent the value of p to determine significance? I know there are probably fixes to the error but since the data is weighted and larger it might be a bit complicated just to get one number.
Below is what I get when I run a table on the weighted dataframe.
table( NM.svy$variables[ c( 'RaceEth' , 'BFDur' ) ] , useNA = 'always' )
This is usually what I get when the chi-square is produced.
![table( NM.svy$variables[ c( 'RaceEth' , 'BFDur' ) ] , useNA = 'always' )](https://i.sstatic.net/GgKlL.png){kind=link}
Design-based Wald test of association data: svychisq(~RaceEth + BF5EVER, IL.svy, statistic = "Wald") F = 10.733, ndf = 5, ddf = 1203, p-value = 4.124e-10
Thank you so much for the answers. After running the table I realized the issue was that the responses weren't evenly distributed. So, I recoded the outcomes to only those I wanted, throwing the rest into an "other" category and that worked!