Do you have a reference to that? I thought it would quietly drop those rows of the data with missing values in fields in the formula. Which is handling missing data, but maybe not in the way you think.

Regards,

Ron.

Do you have a reference to that? I thought it would quietly drop those rows of the data with missing values in fields in the formula. Which is handling missing data, but maybe not in the way you think.

Regards,

Ron.

You have 68 participants (PPT) in one intervention, and 53 in the other, giving a total of 121. I think elsewhere you describe your data as 121 rows with 13 columns. So do you only have one observation per level of PPT? If so, this should be the source of your problem - residual effects being completely confounded with the PPT effect. In this case I think you want an lm rather than an lmer.

Or are participants reused either within or across interventions in some way?

Ron.

Hi Ron,

Thank you for your response.

I have a total of 121 participants in a between-subjects design.

Yes, I only have one observation of each variable per level of PPT

The lmer function is one I have previously been advised to use, do you think this is not the case?

It is not possible to fit the model you are trying to fit, ie with random, independent PPT, with only one observation per PPT. In this case there is no difference between the PPT and the random, residual error terms, hence the failure to converge and residual variance of zero (3.302e-11).

Hi Ron,

Thanks again for your response

I understand what you are saying and I wonder is there any way to use this model but fit it in another way? The only reason I ask is because I have been assured this is model I should use.

Thanks, Karen

It would be good to know the reasoning behind the assurance you were given that lmer is a preferred approach in this case.

No. The PPT effect is really looking to account for similarities (covariance) between repeated observations on the same individuals, and you don't have any.

Unless you have some other information which provides information on structured covariances between particiipants. For example, in a genetics context you might have pedigree so you know close relatives are more similar than unrelated people with an expected proportion of their genes in common. But really that would be a somewhat different model.

As @nirgrahamuk says, it would be good to know the reasoning behind that suggestion. What are you hoping that PPT is doing in the model?

for me the funamental question, is why are unique identifiers of the participants in the study, included in the dataset of the model.

is it a) purely for ID purposes for diagnostics and audit

b) the model should be fit to particular individuals, i.e. if you are johnsmith or marysue then we need to give you your own model.

c) like was previously suggested in the thread , if its an issue of some individuals appearing repeatedly in the set, and needing to account for that.

if a) then simply PPT should not be in the lmer formula.

if b) then we need to really discuss the intent of the modelling exercise

c) is apparently not the case as its one sample per individual / PPT

@ron @nirgrahamuk thank you for highlighting this to me, I understand this isn't the model for me and I will speak to my supervisor about this. Do you think a multilevel model would be more appropriate?

This was a multilvel model, but different nomenclature is used in different areas (mixed model, hierarchical model, multilevel model).

I'd start by dropping the PPT random effect and doing multiple regression with your interventions and assorted covariates included.

1 Like

Hi again,

To me the problem comes from that you have 98 observations for 94 levels of PPT under a Gaussian model. That way the residual variance is confounded with the variance of the random effect.

Under non-gaussian models that can be OK but here since there is no constraint on the residual variance that cannot be.

The model does get close to convergence but I would not trust it.

Simple solution: select a single observation per group and drop the PPT random effect.

Then you go back to a simple `lm()`

and convergence should not be an issue.

This will make you loose 4 observations which should not be a big deal!

2 Likes

@KarenMillin, if this solved your issue do not forget to click on "Solve" or something like that to indicate others you no longer need help. ++

1 Like

Thank you all for your help.

I have changed the model being used to analyse the data.

2 Likes