Understanding Society User Support: Issueshttps://iserredex.essex.ac.uk/support/https://iserredex.essex.ac.uk/support/support/favicon.ico?15995719382022-01-24T13:17:02ZUnderstanding Society User Support
Redmine Understanding Society User Support - Support #1637 (Feedback): Religion in Covid surveyhttps://iserredex.essex.ac.uk/support/issues/16372022-01-24T13:17:02ZSamir Sweida-Metwally
<p>Hello,</p>
<p>I hope you are well.</p>
<p>I had a question about the Covid survey (adults 16+). Looking at the data I can't seem to locate a variable that denotes religious affiliation, importance of religion, and religious attendance. I was wondering if I could get this data by linking it to the USoc survey. If so, could you point to information on how I could do this and whether it is possible to link both surveys to obtain other information from USoc? If linking is not possible, could you advise how it would indeed be possible to obtain the religion information for all waves of the Covid survey?</p>
<p>As always, thanks a lot for your help & support.</p>
<p>Samir</p> Understanding Society User Support - Support #1257 (Resolved): Weight for unbalanced UKHLS panel ...https://iserredex.essex.ac.uk/support/issues/12572019-10-09T11:02:49ZSamir Sweida-Metwally
<p>Dear Olena,</p>
<p>I hope this message finds you well.</p>
<p>I am undertaking a longitudinal data analysis using wave 1 to wave 8 of UKHLS (I am using all waves - i.e., wave 1, 2, 3, 4, 5, 6, 7 & 8). The model is a two level logistic regression with observations (level 1) nested within individuals (level 2). My focus is on individuals (indresp file) who are aged between 16-64 and who completed a full interview (ivfio ==1 ). I am using Stata 15.1 to run my multilevel model. My dataset is unbalanced.</p>
<p>As I understand it I can only use the weights UKHLS provide to undertake analysis on a balanced panel dataset.</p>
<p>However, I don't want to run my analysis on a balanced panel dataset - <strong>but instead I want to run my model on an unbalanced dataset.</strong> This is because by using a balanced panel dataset I end up losing nearly 2/3rd of my observations and given the sub-groups I am interested in I am ending up with too few observations which is generating imprecise coefficients and too large standard errors/confidence intervals.</p>
<p>I have gone through the online discussion forum, notably the exchange you had with Ewan 4 years ago (<a class="external" href="https://iserswww.essex.ac.uk/support/issues/414">https://iserswww.essex.ac.uk/support/issues/414</a>), and I wanted to make sure I understood your advice correctly.</p>
<p>My questions are as follows:</p>
<p>1) Have I understood correctly that a possible remedy to my situation (<strong>i.e. that I want to run a weighted model on an unbalanced dataset rather than on a balanced dataset</strong>) is that I can use the cross-section weight of the earliest wave in my dataset (in this case wave 1) and apply that weight to all pidps across all waves?</p>
<p>2) If so, am I right that this weight would be <strong>'a_indinus_xw' given that I am looking at wave 1 to wave 8?</strong> That means that if I was looking at wave 2 to wave 8, the weight to be applied would be 'b_indinub_xw'. Is that correct?</p>
<p>3) If my understanding of point 2) (above) is correct, <strong>am I right that 'a_indinus_xw' is the only weight I have to use in my model</strong>? That is, that I don't have to use the cross-section weight for each and every single wave (i.e., waves 2, 3, 4, 5, 6, 7, and 8).</p>
<p>4) In the exchange you had with Ewan (see link above) you spoke of 'scaling the data'. <strong>Am I correct in understanding that scaling does not apply to me</strong> as I am not combining UKHLS with BHPS but only using UKHLS? That means that I would use the cross-section wave (e.g., 'a_indinus_xw' assuming my dataset is looking at wave 1 to wave 8) as is in my model without making any modifications to it, is that right?</p>
<p>5) Am I right in understanding that even if I did use the cross-section weight (e.g., 'a_indinus_xw' assuming my dataset is looking at wave 1 to wave 8) to run a weighted model on an unbalanced dataset, I would be increasing the number of observations in my model relative to running a weighted model on a balanced dataset using the UKHLS provided weight (i.e., 'h_indinus_lw' in this case), but I would still have observations being omitted from my model because any person who responded for the first time after wave 1 (i.e., in wave 2, 3, 4, 5, 6, 7 or 8) would have a weight of '0' as they would not have appeared in 'a_indinus_xw' ?</p>
<p>6) If my undersatidng of point 5) (above) is correct, does that mean that the only way for me to run a weighted model taking into account all the observations in my unbalanced dataset would be to create my own weights?</p>
<p>7) Finally, am I correct that the weight I should use <strong>if I wanted to run a weighted version of my model on a balanced dataset (wave 1 to 8) I would use the weight called 'h_indinus_lw' provided by UKHLS?</strong> Am I correct in understanding that by using the weight provided by UKHLS (here 'h_indinus_lw') I am only able to run a weighted version of my model on a balanced dataset because those pidps that do not have an observation in each and every wave (i.e., wave 1, 2, 3, 4, 5, 6, 7 & 8 in my case) would be given a weight of zero?</p>
<p>Apologies for the long message but I wanted to make sure I provided you with all the information you might need to be able to answer my questions. That said, if I have missed something and you do require further information please do not hesitate to let me know.</p>
<p>Thank you very much for your help Olena.</p>
<p>I look forward to hearing from you.</p>
<p>With very best wishes,</p>
<p>Samir</p>