 |
 |
If you have only 11 time points then recent advice is to model this as a fixed effect (using dummy variables) or to use a Bayesian multilevel model to put a prior on the level 2 variance. see http://link.springer.com/article/10.1007/s10648-014-9287-x The former can be handled in SPSS (but personally I would hate to!). Another approach - which may be preferable for certain research questions - is the relational event model. Not possible in SPSS but available in a few R packages. e.g., see https://www.jstatsoft.org/article/view/v064i05 <https://www.jstatsoft.org/article/view/v064i05>This is more suitable if you are focusing on temporal or ordinal dynamics (e.g., perhaps on repeated patterns of offending). Fitting the models requires more work as the technique is more novel and implementations are less user-friendly. <http://link.springer.com/article/10.1007/s10648-014-9287-x> Thom ________________________________ From: Research of postgraduate psychologists. <[log in to unmask]> on behalf of Jeremy Miles <[log in to unmask]> Sent: 31 March 2017 02:13 To: [log in to unmask] Subject: Re: Analysis query... I would do either a cross-lagged model or a fixed effects regression. These answer slightly different questions, but are good for a small number of time points (and I think you want the cross lag). To do this you need to use SEM software, which isn't super straightforward, but if you're doing Poisson mixed models, you don't have a problem. I wrote a paper using this: http://www.drugandalcoholdependence.com/article/S0376-8716(07)00365-1/abstract<https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.drugandalcoholdependence.com%2Farticle%2FS0376-8716(07)00365-1%2Fabstract&data=01%7C01%7Cthomas.baguley%40NTU.AC.UK%7C08ebeef3e6fb421514ee08d477d328fc%7C8acbc2c5c8ed42c78169ba438a0dbe2f%7C1&sdata=KrnqCTh%2FcGFB81jMHSa3o5qzvyaj%2FPYb%2FMtCoBoJHFU%3D&reserved=0> but there are lots of others you can find. (Sorry to be brief, taking a break from dealing with crises [which you can read about in your favorite newspaper :) ]). Reply if you need more help, and then remind me when I forget. Jeremy On Thu, 30 Mar 2017 at 15:56 Laura Scurlock-Evans <[log in to unmask]<mailto:[log in to unmask]>> wrote: Hello, I wondered if I could ask some advice on an analysis strategy for a complex dataset? I need to analyse a longitudinal dataset which contains information collected over a 15-year period on different types of crime experienced by people of different ages. The dataset is organised by age at which events have occurred to individuals (i.e. all crime types experienced between the ages of 4-7 years of age, 8-11 years of age, etc.). Some individuals have entries for each age-band, but others do not (because they have not reached that age, etc.). Therefore, some respondents will have what appear to be missing data. I need to explore which factors predict whether someone will experience two particular types of crime at each age-point, from the types of crime they have experienced up to that age-band – and within that age-band. I also need to take into account demographic characteristics, such as gender. In the future I may also need to incorporate measures of regional social deprivation, etc. I am most familiar with SPSS in terms of statistical software and so far, I have been exploring two types of analysis to see if they would allow me to answer the questions above: 1) multilevel modelling (whereby levels are age-bands) using Poisson regression to model counts of the occurrence of the two types of crime, and 2) time-series analysis. I particularly want to be able to predict whether experiencing the two crime-points in one age-band predicts experiencing them in a later age-band. Both types of analysis are fairly new to me, and I am not sure if they would allow me to answer the questions I need to answer (or at least explore!). For example, I am unsure whether I have enough levels (e.g. 11 age-bands) for multilevel modelling or time-points for time-series analysis (I read in Tabachnick and Fidell that 50 time-points or more are preferable). I wondered if anyone could say whether they feel if the types of analysis I've been looking at makes any sense, or if they could suggest any other approaches which might be better? I would be very grateful for any advice anyone could offer! Many thanks. Best wishes, Laura. -- -- My employer has nothing to do with this email. DISCLAIMER: This email is intended solely for the addressee. It may contain private and confidential information. If you are not the intended addressee, please take no action based on it nor show a copy to anyone. In this case, please reply to this email to highlight the error. Opinions and information in this email that do not relate to the official business of Nottingham Trent University shall be understood as neither given nor endorsed by the University. Nottingham Trent University has taken steps to ensure that this email and any attachments are virus-free, but we do advise that the recipient should check that the email and its attachments are actually virus free. This is in keeping with good computing practice.