1
$\begingroup$

Advice please.

I have received repeat measures from individuals over time asking them to complete self rating questionnaire

Scoring: rate of self harm (0-10) Severity of self harm (0-10) Rate of suicidal thinking (0-10) Rate of suicidal planning (0-10)

What is the best means of analysis for this.

I have corresponding gender and age for the above.

I have the ability to pool all data points and analyse the relationship between variables

And analyse over time.

I'm getting so very confused. (I am a clinician and not a natural statistitian!)

The measures are taken from people who are accessing support via a service for the above issues so we would hope to see change over time. I would like to look at the rate of effectiveness of the intervention but more so, I would like to examine how the variables behave.

It would appear that people recover, from a crude measures perspective, but that whilst some things obviously improve, others don't.

I'm pretty keen to look at what the variable interplay is.

When I'm a little more confident with the data I want to know at various time intercept points, across individuals, what the variable patterns are over the (up to 2 year intervention)

However, in the first instance I want to understand the relationship between the Self Harm and suicide variables. And how to undertake that analysis. Any help welcomed #outofmydepth

Improve this question
$\endgroup$
10
  • $\begingroup$ What is the nature of the scientific question? No statistical jargon, please. $\endgroup$
    – AdamO
    Feb 10, 2017 at 20:56
  • $\begingroup$ @AdamO I want to understand the relationship between self harm and suicidality. $\endgroup$
    – Caroline Harroe
    Feb 10, 2017 at 21:14
  • $\begingroup$ Can you just do repeated measures ANOVA with blocking factors of age / gender? You will probably need to consider effect modification by gender. I don't understand how a longitudinal aspect would play into it unless there is interest in growth effects but it doesn't seem to be part of your hypothesis. $\endgroup$
    – AdamO
    Feb 10, 2017 at 21:20
  • 1
    $\begingroup$ Caroline, please edit the question to include ALL the pertinent details. An intervention is a rather large omission. Also hashtags do not work on this site, you should use the tag feature which you already have. $\endgroup$
    – AdamO
    Feb 10, 2017 at 21:55
  • 1
    $\begingroup$ Frankly no, your description does not make sense. I think for this project you will need access to a statistical consultant. $\endgroup$
    – AdamO
    Feb 11, 2017 at 15:53

2 Answers 2

Reset to default
1
$\begingroup$

I agree with others that it sounds like you may need someone to help you analyze this data, because what I am going to suggest is not likely something that can be undertaken by a novice. Having said that - depending how many repeated measures you have, a good approach for this data may be longitudinal hierarchical linear modeling (HLM). Conceptually, HLM allows you to estimate all the effects it seems like you are interested in 1) the trajectories of change for your variables, for both individuals and in aggregate (HLM allows for the estimation of the 'average; change trajectory and also the variability of individuals around that average change); 2) the effects of person level variables on the individual change trajectories, for example the effects of sex or age; and 3) the effects of a time-varying covariate (suicidality over time) on your outcome (I am not clear about your hypothesis so I am guessing that what you mean is how changes in one variable effect the observed changes in the other). HLM would do all these things for you, and also allow you to ask questions about non-linear change in your outcome. HTH.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ This is really helpful, thankyou. And yes, precisely what you have said. There are multiple points, fortnightly for up to two years. Although there is variance in the number of repeat measures the mean number of repeat measures is 12.5. I am keen to find a way to do this well and on limited resources that will be tough, but if I know where to start I can learn. What is important is to know whether the self harm scales impact the suicidality (or conversely) and whether this interaction varies over time. $\endgroup$
    – Caroline Harroe
    Feb 11, 2017 at 19:36
0
$\begingroup$

When a trial collects more than one outcome--and here I am not referring to repeated measures, but proxies, items, or imperfect measures or responses with measurement errors--there are many important considerations and ways to analyze those measures. You allude to the correlation between two or four items in the outcome. However, you mention later that you have an intervention. To appropriately analyze the outcome, you must settle first on the primary analysis. This is because there will naturally be changes in the outcome, but the validity of those measures must account for your experimental manipulation of the subjects' conditions.

It seems you have not settled on a primary analytic approach. @Marina_ANOVA suggests HLM and I echo this suggestion: it is a standard approach to evaluation of trials with an outcome of a psychometric nature such as responses to a questionnaire or a battery of cognitive or emotional tests. An HLM can be estimated with structural equation modeling or with random slopes models. This is a powerful approach because it treats several different outcomes as random effects with common slope term. Therefore your original question about the correlation between them is moot. HLMs exploit such correlation or even a lack thereof, to find trends among the underlying sociologic construct: suicidality in your example. The correlation can be determined by looking at path weights or effects relating your latent variable "suicidality" to its many exogenous measures.

The following paper from a colleague of mine has an introduction which may be an illuminating read: http://onlinelibrary.wiley.com/doi/10.1002/sim.7035/full

Statistical software M-plus can fit these models, which is appropriate even for a novice. See examples here: M-plus http://statistics.ats.ucla.edu/stat/mplus/seminars/gm/ SAS: http://www.ats.ucla.edu/stat/sas/seminars/sas_mlm/mlm_sas_seminar.htm Stata: http://www.ats.ucla.edu/stat/stata/faq/growth_xtmixed_sem.htm

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.