[R-sig-ME] longitudinal analysis when one group switched from control to treatment

Wed May 20 17:22:54 CEST 2020

Please keep the mailing list in cc

1) Both boxplots display the same information. Choose the one you find the
most clear.
2) In an ideal world, all boxplots should look similar. Determining if they
are sufficiently similar is the point where statistics becomes an art
instead of science.
3) You could consider a random effect of the year-school interaction
(1|scid:year). This has the downside that each year-school combination gets
an independent estimate, ignoring a common school effect. It's up to you to
compare an improvement in fit vs the interpretability of the model.

Best regards,

Thierry

ir. Thierry Onkelinx
Statisticus / Statistician

Vlaamse Overheid / Government of Flanders
INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
FOREST
Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
thierry.onkelinx using inbo.be
Havenlaan 88 bus 73, 1000 Brussel
www.inbo.be

///////////////////////////////////////////////////////////////////////////////////////////
To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey
///////////////////////////////////////////////////////////////////////////////////////////

<https://www.inbo.be>

Op di 19 mei 2020 om 15:26 schreef Simon Harmel <sim.harmel using gmail.com>:

> Dear Thierry,
>
> Thank you very much for the demonstration. Three questions:
>
> (1) Are two different sets of boxplots needed?
> (2) From these boxplots, how can we exactly determine the the failure of "
> m2" in capturing the switch of a school from control to treatment (what
> are the criteria)?
> (3) Suppose "m2" failed, what is generally an alternative to account for
> the switch  of a school from control to treatment?
>
> Thank you, Simon
>
> On Tue, May 19, 2020 at 2:27 AM Thierry Onkelinx <thierry.onkelinx using inbo.be>
> wrote:
>
>> Dear Simon,
>>
>> Something like this
>>
>> library(lme4)
>> dat <- read.csv('https://raw.githubusercontent.com/hkil/m/master/z.csv')
>> dat$year <- factor(dat$year)
>> m2 <- lmer(y~ year*group + (1|scid/stid), data =dat)
>> dat$resid <- resid(m2)
>> library(ggplot2)
>> ggplot(dat, aes(x = year, y = resid, colour = group)) +
>>   geom_boxplot() +
>>   facet_wrap(~scid)
>> ggplot(dat, aes(x = scid, y = resid, colour = year, fill = group)) +
>>   geom_boxplot()
>>
>> Best regards,
>>
>> ir. Thierry Onkelinx
>> Statisticus / Statistician
>>
>> Vlaamse Overheid / Government of Flanders
>> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE
>> AND FOREST
>> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
>> thierry.onkelinx using inbo.be
>> Havenlaan 88 bus 73, 1000 Brussel
>> www.inbo.be
>>
>>
>> ///////////////////////////////////////////////////////////////////////////////////////////
>> To call in the statistician after the experiment is done may be no more
>> than asking him to perform a post-mortem examination: he may be able to say
>> what the experiment died of. ~ Sir Ronald Aylmer Fisher
>> The plural of anecdote is not data. ~ Roger Brinner
>> The combination of some data and an aching desire for an answer does not
>> ensure that a reasonable answer can be extracted from a given body of data.
>> ~ John Tukey
>>
>> ///////////////////////////////////////////////////////////////////////////////////////////
>>
>> <https://www.inbo.be>
>>
>>
>> Op ma 18 mei 2020 om 19:35 schreef Simon Harmel <sim.harmel using gmail.com>:
>>
>>> Dear Thierry,
>>>
>>> Would you possibly demonstrate what you exactly mean perhaps in R?
>>>
>>> Thank you,  Simon
>>>
>>> On Mon, May 18, 2020, 12:13 PM Thierry Onkelinx <
>>> thierry.onkelinx using inbo.be> wrote:
>>>
>>>> Dear Simon,
>>>>
>>>> You are wondering if the model captures the change in treatment for a
>>>> school. Therefore you need to plot the residuals vs every combination of
>>>> school and year. A boxplot for every combination would be useful. If the
>>>> change in treatment triggers a shift in residuals, then the current model
>>>> fails.
>>>>
>>>> Best regards,
>>>>
>>>> ir. Thierry Onkelinx
>>>> Statisticus / Statistician
>>>>
>>>> Vlaamse Overheid / Government of Flanders
>>>> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE
>>>> AND FOREST
>>>> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
>>>> thierry.onkelinx using inbo.be
>>>> Havenlaan 88 bus 73, 1000 Brussel
>>>> www.inbo.be
>>>>
>>>>
>>>> ///////////////////////////////////////////////////////////////////////////////////////////
>>>> To call in the statistician after the experiment is done may be no more
>>>> than asking him to perform a post-mortem examination: he may be able to say
>>>> what the experiment died of. ~ Sir Ronald Aylmer Fisher
>>>> The plural of anecdote is not data. ~ Roger Brinner
>>>> The combination of some data and an aching desire for an answer does
>>>> not ensure that a reasonable answer can be extracted from a given body of
>>>> data. ~ John Tukey
>>>>
>>>> ///////////////////////////////////////////////////////////////////////////////////////////
>>>>
>>>> <https://www.inbo.be>
>>>>
>>>>
>>>> Op ma 18 mei 2020 om 17:45 schreef Simon Harmel <sim.harmel using gmail.com>:
>>>>
>>>>> Dear Thierry,
>>>>>
>>>>> By "Have a look at the residuals" you mean something like the
>>>>> following (below)? So no other adjustment is required for the switching
>>>>> that occurred?
>>>>>
>>>>> plot(m1, type = c("p","smooth"), col.line = 2)
>>>>>
>>>>> plot(m1, sqrt(abs(resid(.)))~fitted(.), type = c("p","smooth"),
>>>>> col.line = 2)
>>>>>
>>>>> On Mon, May 18, 2020 at 2:01 AM Thierry Onkelinx <
>>>>> thierry.onkelinx using inbo.be> wrote:
>>>>>
>>>>>> Dear Simon,
>>>>>>
>>>>>> The question is rather if the model is able to capture this change.
>>>>>> Have a look at the residuals. If they look OK, then the model handles the
>>>>>> change in treatment.
>>>>>>
>>>>>> Best regards,
>>>>>>
>>>>>> ir. Thierry Onkelinx
>>>>>> Statisticus / Statistician
>>>>>>
>>>>>> Vlaamse Overheid / Government of Flanders
>>>>>> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR
>>>>>> NATURE AND FOREST
>>>>>> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
>>>>>> thierry.onkelinx using inbo.be
>>>>>> Havenlaan 88 bus 73, 1000 Brussel
>>>>>> www.inbo.be
>>>>>>
>>>>>>
>>>>>> ///////////////////////////////////////////////////////////////////////////////////////////
>>>>>> To call in the statistician after the experiment is done may be no
>>>>>> more than asking him to perform a post-mortem examination: he may be able
>>>>>> to say what the experiment died of. ~ Sir Ronald Aylmer Fisher
>>>>>> The plural of anecdote is not data. ~ Roger Brinner
>>>>>> The combination of some data and an aching desire for an answer does
>>>>>> not ensure that a reasonable answer can be extracted from a given body of
>>>>>> data. ~ John Tukey
>>>>>>
>>>>>> ///////////////////////////////////////////////////////////////////////////////////////////
>>>>>>
>>>>>> <https://www.inbo.be>
>>>>>>
>>>>>>
>>>>>> Op zo 17 mei 2020 om 01:09 schreef Simon Harmel <sim.harmel using gmail.com
>>>>>> >:
>>>>>>
>>>>>>> Hello All,
>>>>>>>
>>>>>>> I have a 3-year longitudinal dataset (*see link below the table*).
>>>>>>> Up to
>>>>>>> year 2 (coded "1"), 8 schools (4 in Treatment, 4 in Control)
>>>>>>> cooperated
>>>>>>> with the study. But in year 3 (coded "2"), one of the Treatment
>>>>>>> schools
>>>>>>> (named "good") dropped out.
>>>>>>>
>>>>>>> Also in year 3 (coded "2"), we were made to move one of the *Control
>>>>>>> *schools
>>>>>>> (named "*orange*") to the *Treatment *group. The full design of the
>>>>>>> study
>>>>>>> is shown in the Table below.
>>>>>>>
>>>>>>> I want to regress "year" and "group" on "y" (a continuous response)
>>>>>>> in lme4
>>>>>>> package in R. But is there a way to capture the switch of one of the
>>>>>>> control schools to the treatment group?
>>>>>>>
>>>>>>> Thank you very much, Simon
>>>>>>>
>>>>>>> ·       *Switched from control to treatment*
>>>>>>>
>>>>>>> ·       *Out as of year coded 2*
>>>>>>>
>>>>>>> *SCHOOL NAMES*
>>>>>>>
>>>>>>> *Year*
>>>>>>>
>>>>>>> *Codes*
>>>>>>>
>>>>>>> *Control*
>>>>>>>
>>>>>>> *Treatment*
>>>>>>>
>>>>>>> 0
>>>>>>>
>>>>>>> har
>>>>>>>
>>>>>>> john
>>>>>>>
>>>>>>> bus
>>>>>>>
>>>>>>> orange
>>>>>>>
>>>>>>> caro
>>>>>>>
>>>>>>> good
>>>>>>>
>>>>>>> bla
>>>>>>>
>>>>>>> carm
>>>>>>>
>>>>>>> 1
>>>>>>>
>>>>>>> har
>>>>>>>
>>>>>>> john
>>>>>>>
>>>>>>> bus
>>>>>>>
>>>>>>> *orange*
>>>>>>>
>>>>>>> caro
>>>>>>>
>>>>>>> good
>>>>>>>
>>>>>>> bla
>>>>>>>
>>>>>>> carm
>>>>>>>
>>>>>>> 2
>>>>>>>
>>>>>>> har
>>>>>>>
>>>>>>> john
>>>>>>>
>>>>>>> bus
>>>>>>>
>>>>>>> X
>>>>>>>
>>>>>>> caro
>>>>>>>
>>>>>>> *orange*
>>>>>>>
>>>>>>> bla
>>>>>>>
>>>>>>> carm
>>>>>>>
>>>>>>> *library(lme4)*
>>>>>>> *dat <- read.csv('
>>>>>>> https://raw.githubusercontent.com/hkil/m/master/z.csv
>>>>>>> <https://raw.githubusercontent.com/hkil/m/master/z.csv>')*
>>>>>>>
>>>>>>> *m1 <- lmer(y~ year*group + (1|stid), data = dat)      #### 'stid' =
>>>>>>> student id                m2 <- lmer(y~ year*group + (1|scid/stid),
>>>>>>> data =
>>>>>>> dat) #### 'scid' = school id*
>>>>>>>
>>>>>>>         [[alternative HTML version deleted]]
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> R-sig-mixed-models using r-project.org mailing list
>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>>>>>
>>>>>>

	[[alternative HTML version deleted]]