''TIME SERIES'' ANALYSIS

[researcher]

Supposed i have a survey N,1000 nationally representative on the usual demographics

I  carry out a new wave of the survey every 3 months with a new sample but the questions are constant.

I currently have 4 waves of this survey.

Now Suppose I want to show whether the change in a particular question (say support for a particular politic party, on a  likert scale)  is ''significant'' over time? Say from wave 1 to wave 2 or even comparing wave 1 and 4).

What tests / approaches are best for this?

Is it even a meaningful question?

All thoughts appreciated. This is hypothetical but a colleague asked  me about a similar issue and I could not think what to advise.

The samples in the different waves would be 'unrelated' so it should be about comparing two means (for any two time points?) so perhaps t test for independent samples or non parametric equivalent (Mann Whitney) .

Would it be possible to compare 3 or more time points using ANOVA and post hoc tests.

I have a feeling I am opening a can of worms here but hey ho..

 

[Maguin, Eugene]

Anova with post-hocs to test all possible comparisons or contrasts to test specific (planned) comparisons.
Gene Maguin


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of researcher
Sent: Wednesday, November 16, 2016 4:55 PM
To: [hidden email]
Subject: ''TIME SERIES'' ANALYSIS

Supposed i have a survey N,1000 nationally representative on the usual demographics

I  carry out a new wave of the survey every 3 months with a new sample but the questions are constant.

I currently have 4 waves of this survey.

Now Suppose I want to show whether the change in a particular question (say support for a particular politic party, on a  likert scale)  is ''significant'' over time? Say from wave 1 to wave 2 or even comparing wave
1 and 4).

What tests / approaches are best for this?

Is it even a meaningful question?

All thoughts appreciated. This is hypothetical but a colleague asked  me about a similar issue and I could not think what to advise.

The samples in the different waves would be 'unrelated' so it should be about comparing two means (for any two time points?) so perhaps t test for independent samples or non parametric equivalent (Mann Whitney) .

Would it be possible to compare 3 or more time points using ANOVA and post hoc tests.

I have a feeling I am opening a can of worms here but hey ho..

 





--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/TIME-SERIES-ANALYSIS-tp5733476.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Maurice Vergeer]

this is what you would call "repeated cross sectional surveys".

Panel design uses the same respondents, while time series analysis has at least 50 points in time.

Because you state these are different respondents, it makes analysis easier because observations are/should be uncorrelated across time.

Crosstabulations, anova, regression analysis are options for analysis.

The green book series by Sage on methods and design has one specifically about repeated cross sectional surveys.

Hope that helps.

Maurice

On Wed, Nov 16, 2016 at 10:55 PM, researcher <[hidden email]> wrote:

Supposed i have a survey N,1000 nationally representative on the usual
demographics

I  carry out a new wave of the survey every 3 months with a new sample but
the questions are constant.

I currently have 4 waves of this survey.

Now Suppose I want to show whether the change in a particular question (say
support for a particular politic party, on a  likert scale)  is
''significant'' over time? Say from wave 1 to wave 2 or even comparing wave
1 and 4).

What tests / approaches are best for this?

Is it even a meaningful question?

All thoughts appreciated. This is hypothetical but a colleague asked  me
about a similar issue and I could not think what to advise.

The samples in the different waves would be 'unrelated' so it should be
about comparing two means (for any two time points?) so perhaps t test for
independent samples or non parametric equivalent (Mann Whitney) .

Would it be possible to compare 3 or more time points using ANOVA and post
hoc tests.

I have a feeling I am opening a can of worms here but hey ho..







--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/TIME-SERIES-ANALYSIS-tp5733476.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

--

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[researcher]

Thanks for all feedback - not had a chnace to work through it in detail yet - much appreciated though !

[researcher]

Thanks for these massively helpful replies which I am following up.

[Ryan]

Dear _____,

It's possible that if you sample from the same geographic areas (e.g. x number of residents sampled from each of y number of zip codes per time period), accounting for residual correlation within the same geographic area over time may warranted.

Ryan

Sent from my iPhone

> On Nov 16, 2016, at 4:55 PM, researcher <[hidden email]> wrote:
>
> Supposed i have a survey N,1000 nationally representative on the usual
> demographics
>
> I  carry out a new wave of the survey every 3 months with a new sample but
> the questions are constant.
>
> I currently have 4 waves of this survey.
>
> Now Suppose I want to show whether the change in a particular question (say
> support for a particular politic party, on a  likert scale)  is
> ''significant'' over time? Say from wave 1 to wave 2 or even comparing wave
> 1 and 4).
>
> What tests / approaches are best for this?
>
> Is it even a meaningful question?
>
> All thoughts appreciated. This is hypothetical but a colleague asked  me
> about a similar issue and I could not think what to advise.
>
> The samples in the different waves would be 'unrelated' so it should be
> about comparing two means (for any two time points?) so perhaps t test for
> independent samples or non parametric equivalent (Mann Whitney) .
>
> Would it be possible to compare 3 or more time points using ANOVA and post
> hoc tests.
>
> I have a feeling I am opening a can of worms here but hey ho..
>
>
>
>
>
>
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/TIME-SERIES-ANALYSIS-tp5733476.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD