cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Choose Language Hide Translation Bar
Easy DOE: Easy Enough for a Seven-Year-Old? (2022-US-30MP-1163)

Ryan Lekivetz, JMP Principal Research Statistician Developer, SAS
Rory Lekivetz, First grade student, Northwoods Elementary School

 

JMP 17 introduces the Easy DOE platform, providing both flexible and guided modes to users, aiding them in their design choices. In addition, Easy DOE allows for the DOE workflow from design through data collection and modelling. This presentation offers a preview of the new Easy DOE platform, including insights from a seven-year-old using the new platform on a classic DOE problem.

 

 

Hello. Thank  you  for  joining  us  today.

I'm  Ryan  Lekivetz,  Manager  of  the  DOE and  Reliability  team  at  JMP.

And  I'm  Rory  Lekivetz, rising  second- grade  student.

Today ,  we're  going to  talk  to  you  about  Easy  DOE.

I  posed  the  question easy  enough  for  a  seven- year- old.

Now,  for  a  lot  of  you  watching, you may  not  have  heard  of  Easy  DOE  before.

It's  one  of  our  new  platforms  in  JMP 17,

and  so  it's  base  JMP so  you  don't  need  a  pro  to  use  it.

The  idea  with  Easy  DOE  is  that we're going  to  have  one  file,

one  workflow  that's  going  to  contain   both  the  design  and  analysis.

If  you're  familiar  with  doing   design experiments and JMP,

you're  used  to  going   under  the  DOE  menu,

creating  a  designed  experiment, and  then  making  a  data  table  from  there.

And  then  you  would  do  all   your  data  table  on  that  analysis.

And  there  was   that  separation between  the  design  and  analysis  point.

The  idea  with  Easy  DOE  is  that

we're trying  to  aid  novice  users through  the  entire  workflow.

And  so,  unlike  custom  design, you're  going  to  see  a  lot  more  hints

and  different  defaults that are set  to  try  to  aid  those  users.

And  you're  going  to  see  that  both on  the  design  and  the  analysis  side.

In addition,  you'll  also  see   there  isa flexible mode

for  those  who  are   more  comfortable  with  DOE.

That's  all  safe.

My  purpose  in  this  talk   was  really to see,

is  Easy  DOE  going to  be  easy  enough  to  use?

Well,  I  said  seven- year- old,

but  I  guess  you're  more  like seven and  a  half  now,  is  that  right?

The  idea  here  was  to  let  Rory   do the steering throughout.

That  I  wanted  her  to  be  the  one   using  Easy  DOE,  putting  everything  in.

I  wanted  to  have   as  little  input  as possible,

even  when  it  came  to  decisions   about  what to do with the design,

and  things  like  that.

Of  course,  I  still  did  need to  give  her  an  introduction  to  DOE.

I  mean,  if  you've  seen in our DOE documentation,

this  figure  might  look  familiar.

So  we  went  through the  different  phases.

We  talked  about  what's  the difference between a goal,

a  response  and  the  factors.

Specify.

That's  where  we  talked  about   what's  going  on  with  the  model?

And  in  particular,  what's  the  difference between a  main  effect  and  an  interaction?

Why  would  you  care   about  one  or  another?

And  we  said,  well,   once  we  have  the  design,

we  actually  have  to  go   about  collecting some data.

Then  we  have  to  fit,  then  we have  to  do  something  with  the  model.

We  want  to  find  out  what's  important

and  how  can  we  use  that  model to  do  something  further.

I  mean,  you  can  imagine  we  spent maybe about half an hour

just  talking  about   some  of  those  different  things.

I'll  say,  well,  why  don't  we  use some  classic  experiments ?

My  suggestions  were,  why  don't  we  try the  old  paper  helicopter  experiment?

O n  our  shelf  we  have  the  statap ult or  the  classic  catapult  experiment.

And  so  what  did  Rory  say?

No.

Rory's  idea  was  actually to  do  paper  airplanes.

S he'd  started  to  to  try  out flying  some  paper  airplanes.

S he  said,  well,  I  want  to  do a  DOE  with  paper  airplanes.

I  said,  well,  that's  great, let's  see  what  we  can  do.

Luckily,  she  knew   the  classic  paper airplane

which  is  what  you'll  see is  called  a  Dart.

But  we  found  a  website   that  had  different instructions

for that different  types of  paper  airplanes  you  could  make.

And  on  top  of  that,  it had suggestions

for  what  you  could  do to  try  to  make  your  plane  fly  better.

So  thankfully  then, instead of her  having to try to figure out

what  are  some  of  her  factors   and  levels going to be,

this  website  had  some really  nice  suggestions  for  that.

Before  we  get  into  this, Easy  DOE,   as  I  said, it is that new platform,

and  you're  just  going  to  find  it   under  the  DOE  menu.

Underneath  custom  and  augment  design, there's  Easy  DOE  right  there.

Now,  what  you'll  see   with  Easy  DOE  is that idea

of  going  through  that  workflow is  going  to  be  done  via  tabs.

A s  soon  as  you  launch Easy  DOE,

you'll  see  there's  that   guided  and  flexible  mode.

We're  just  talking  about   guided  mode here.

But  the  idea  is  we're  going  to  go

through these  different  steps by  clicking  on  the  tabs.

There's  the  Define,

and  then  we're  going  to  go to  the  Model,  Design,  et  cetera.

One  way  to  do  that  is  to  click on  the  tabs,  the  tabs  one  at  a  time.

And  at  the  bottom  of  Easy  DOE,

you'll  also  find  a  set of  navigation controls

so  that  will  take  you  forward  and  backwards  between  the  different  tabs.

And  so  the  idea  of  our  talk  here,

we're  going  to  go  through   these  different tabs

and  both  of  us  are  going   to  give  observations.

Rory  is  going  to  give  her  thoughts first on those different tabs

and  then  followed  by  my  own.

Think  of  it more  as  a  teacher in  my  point  of  view,

and  Rory's  was  more as  the  novice  user, trying  Easy  DOE  for  the  first  time.

Let's  start  with  the  Define  tab.

On  this  one,  we  had  to  type  in   the  different  factors and levels

that  we  were  going to  do  for  a  paper  airplane.

None  of  my  factors  had  numbers, so I chose categorical.

The  different  levels   and  factors  that  I  had  type,

and  then  the  levels   were  dart  and  lock bottom.

And  then  there  was  paper, and the levels were  regular  and construction.

In throwing  force, the levels  for  that  one  was  hard and light.

And  paperclip  was  paper  clip   and  no  paper  clip.

The  response  that  we  had   was  distance.

And  for  the  distance,  I  wanted  to  see

what  could  make   the  paper  airplane  go the  farthest

and  our JMP  goal  was  maximize.

Just  to  mention.

When  you  go  into  Easy  DOE, you're going to find right now.

If  you  take  a  look  at   that  screenshot  from  before.

So  currently,  we  have three different  types  of  factors:

continuous,  discrete,  numeric,   and  categorical  factor  types.

I'll  say  I  think she was  able to identify

that  she  needed  the  categorical   for  all  of  those.

Of  course,  now,   she  did  need  confirmation.

I  mean,  when  she  said  categorical,

she  looked  back  to  me  to  see, "Am  I  doing  that  correctly?"

S he  was  able  to  identify  the  factor types and actually enter her level names,

sorry,  her  factor  names   and  her  levels.

Now,  I  will  admit  though, she's  used  to  using  a  touch screen.

And  so  there  did  come  a  point

instead of  her  trying  to  click  into the little box for levels

or  the  factor  name  and to do a  double  click,

there  did  come  a  point  where I told her

you  can  just  use  the  tab  button   to  make  your  life  easier.

A gain,  I  didn't  want  to  have   a  whole  lot  of  input.

S he  picked  the  four  factors  and  levels

and  I  think  I  had  told  her   to  pick  three  or  four  was   a  good  number.

But  I  will  say  so  if  you  think  back  to that paper clip.

She  had  mentioned  it  was   the  paper  clip  or  no  paper  clip,

that  one  took  a  little  bit  of  time.

I  think  maybe  if  we  would  have  went  paper clip yes or no,

it  might  have  been  easier.

But  I  think  for  her,  it  was.

Well,  how  do  we actually define  a paper clip factor

that  also  has  a  paper  clip  level?

But I  think  once  she  had  that  idea, okay, it  was  just  kind  of  a  yes  or  no,

it's  in  there  or  not, we  were  able  to  get  through  it okay.

Moving  on  from  the  Define  tab.

The Model.

For  the  Model  tab,

we  picked  the  one  which   we  thought  would be  the  best.

We decided that Main Effects   and two- factor  interactions

were  the  ones  that   we  want   the number.

So  we  picked  the  one  that  had Main Effects  and  two- factor  interactions.

The  number  of  runs  meant we  had  to  make  16  airplanes

which  didn't  seem  too  bad.

Now,  I'll  say  in  hindsight   for  my  own  view  on  the  Model  tab,

I  mean  this  required  the   most  hand-holding of all the tabs.

And  this  was  really  more  because   of  trying to explain the difference

if  we  just  went  with  main  effects  versus  interactions.

But  again,  so  this  is   a  seven  and  a  half- year- old

who's  never  taken  any statistics course,

has  never  done  any  kind   of  modeling  before.

If  you  have  somebody  who's familiar with the  idea  of  main effects and interactions,

then  I  think  that  tab   wouldn't  be  nearly  as  bad.

Now,  I  will  say  too,  though , it  was  nice.

The  paper  airplane  website   that  we  were using,

it  spelled  out  the  idea

of  what  interactions  really  mean  with  those  factors.

I  think  so  you  would   see  things  where  it  said:

while  some  of  the  airplanes   will  work  better  with  a  paper  clip,

or  you'll  see  some  airplanes  are  better  when they're throwing hard,

while  others  need  that  lighter  touch.

In  some  sense,   that  actually  gave  us

a  natural  point  to  start talking  about  interactions.

Now,  also,  say,  when  I  think  of   from  my  own perspective,

it's  also  something  that  we  can  improve  upon.

From  a  US,  I  want  to  think  about,

well,  how  can  we  distinguish   between  these  choices?

We  did  have  the  hints   in  there  as  well.

But  when  I  think  of,  if  I  wasn't  there to do the hand-holding,

how  might  she'll  come  up   with  that  decision?

Now  when  we  went  to  that  Model  tab,

and  our  next  tab  that  we clicked  on  was...

Design.

There's  not  really  a  lot   I  can  say  about  this  one,

but  I  thought  it  was  interesting that it was showing

what  kind  of  paper  airplanes   that  you  were  going  to  make.

And  then  we  had  the  hard  work of  making  planes  and  flying  them.

Yeah,  I  think  that  probably   took  the  most time, but yeah.

I  don't  have  too  much  to  add  on  this  one.

I  think,  though, it was nice  to have that design displayed

just  to  give  her  that  sense  of,

well,  what  did  that  really  mean when  we  put  in  those  16  runs?

When  we  have  those  factors,   what  does  that  mean  at  the  end  of  the  day?

I  think  from  this, then  she  could  really  get  that  sense.

Okay,  then  my  plane  one, it's  the  lock bottom  with  regular  paper.

I'm  going  to  throw  it  lightly and  put  a  paper  clip  on  it.

What  was  the  next  tab   that  we  went  to?

Data  Entry.

So  full  of  a  data  entry.

We  didn't  really  have  to  do that  much  in  this  one.

So  we  just  put  the  distance  of the  different  kind  of  airplanes  flew.

So  that  was  what  we  measured with  our  measuring  tape  of  stuff.

Is  that  right?

Now,  I'll  say  with  this  one,  if  you  look,   there  were  these  factor  plots.

This  is  just  for  the  main  effects  here,

but  this  actually  was   telling  her   what  she  had  said as soon as she was done.

She  see  those  factor  plots   at  the  bottom,

and  she  said  the  lock bottom   is  not  the  best.

I  think  she  might  have  used   something stronger than that,

but  yeah,  so  the  lock bottom   was  not  good  for  her.

Also,  this  was  a  good  teaching  moment  about randomization.

So  the  run  order  does   come  out  randomized.

I  did  have  to  warn  her  when  we  had all the paper airplanes with us outside,

I  said,  well,   you  don't  want  to

throw all the  dart  type  first, followed  by  the  long  bottom.

I  said,  because  you  might  get  better as you were throwing

or  it  might  start  to  storm,   it  might  get  windier.

Now,  I  say  some  of  these  results.

There  were  times  where  the  hard- throwing   force  needed  a  bit  of  practice.

I'll  admit  there  were  a  few  there that probably were based on more than one,

because  if  you  had  a  crash, almost immediate crash landings.

But  I'll  say  it  did  seem   straightforward for her

to  be  able  to  enter the  data  right  from  there.

I  mean,  I  didn't  have  to  say  anything; she   knew.

So  when  you  come  in  to  hear it,

the  response  column   just  had  missing  values.

And  so  she  had  that  intuition,

well,  this  is  where  I  need  to  go  to  put  in  the  data.

I  also  mentioned  here,  you'll  see this  export  data  and  load  response.

That  export  data  is  if  you  actually

just want  to  create  a  data  table   with  all  your  stuff.

S ometimes  that  will  be  useful   if  you want to go through

what  you're  typically  thinking  of   with  your  JMP  workflow.

If  you  just  want  to  JMP  data  table,

that's  what  that  export  data button  is  going  to  do.

Likewise,  load  response, if  you've  actually  just  recorded

your responses  in  a  different data  table,  you  can  do  that.

Now  that  we  had  our  data, what  did  we  have  to  do?

-What  was  our  next  step? -Analyze.

For the analyze,  I  already  knew   that  the  dotted  ones

weren't  the  most  important  ones.

I  figured  out  I  just  needed to  click  them  to  get  rid  of  them.

I  found  out  dart  type  was one  of  the  most  important  ones.

Now,  I'll  say  here.

The  analyze  I  had  thought  was  going to  be  the  hardest  one  to  explain.

Now,  I'll  say  it  was  surprisingly easy  and  effective.

It  was  almost  she  had  clicked  on  the  tab and  just  started  doing  her  own  thing,

and  I  didn't  really need  to  say  an  awful  lot.

Now,  you'll  notice  here,

so  when  you  come  into  that  analyze  tab, when  Rory  saw  it,  it  was  the  full  model,

but  she  saw  a  lot  of  the  terms   that  were  not  significant.

There  were  a  lot  that  were  dashed and   close  to  that  zero.

And  so  what  she  did  was  just   remove  those  ones  one  at  a  time.

And  then  I  also  see   there  is  a  best model button

that's  based  on  some  type   of  a  forward  selection.

Now,  I'll  say  the  best  model   may actually have more terms,

but  at  the  end,  you  can  see

those extra terms  really  weren't  even  significant.

S ometimes  I  couldn't  even argue  with  the  results.

Perhaps,  one  might  even  argue that  the  model  that  Rory  picked  was  better

because  it  was  simpler  and  there wasn't a  huge  difference  between  the  two.

Now,  one  of  the  nice  things   with  this  Easy DOE platform,

this  analyzed  for  the  guided  users,

there's  this  idea  of  adding and  removing  terms  easily.

And  so  to  add   and  remove  these  terms,

all  that  you  do  is  you  go   to  those confidence intervals in there

and  a  click  will  either  add  it   or  remove it

depending  on  if  it's  currently   in  the  model.

I  highly  recommend  trying  that  out   when you get your hands on 17.

Perhaps,  there  was  something   to  be  said.

The  best  model, when we'll see  in  a  minute with the profiler,

it  makes  it  a  little  bit   more  interesting, of course,

if  you  have  some additional  terms  in  there.

Rory's  model, with  only  the  three  terms,

the  profiler  was  less  interesting to  try  to  explain  things.

But  again,  for  being   a  first  time  DOE,

a  model  like  that   didn't  disappoint  me  at  all.

Now  that  we  had  a  model, what  do  we  actually  do  with  that?

-So  what  was  our  next  step? -Predict.

The Predict tab, this  tab  tells  you what the  best  paper  airplane  was.

If  you  click  on  the  levels   that  you  think are the worst,

it  shows  what  might  happen   with  that  airplane.

Likewise  with  that  Model  tab,

if  you  have  a  user  who's  used  the  profiler before in something like Fit model,

of  course,  that'll  be  a  lot  easier.

But  I'll  also  say the profiler  is intuitive in general.

I  think  it  was  easy   for  her  to  pick  up  on

once  she  started  playing  around   with  that  profiler.

And  then  just   a  little  bit  of  an  idea,

a  discussion  as  to   what  that  actually  meant.

But  I  think  the  profiler,

that's  the  nice  thing  with  the  Easy  DOE, the  profiler  is  already  intuitive.

It  was  also  a  good  teaching  moment   when  it  came  to  interactions.

If  you  go  back,   if  you  look  at  that  model,

we  actually  had  an  interaction between  the  type  and  the  type  of  paper.

Paper  and  type  had  an  interaction and  so  then  we  could  talk  about

well, what happens  when  we  change  type, then look  at  what  happens  with  that  paper?

Again,  this  is  where   those  extra  model terms.

In  Rory's  model,  that  hard  and  light   had  a  zero  effect.

I  was  saying  it  didn't  really  matter what you did for hard or light.

The  best  model,  it  had  a  small  effect,

so  you  could   say,  well,  it's  not  going   to  make  a  big  difference.

But  for  some  of  these, it  was  saying  that  it  was.

But  again,  I  think  the  Predict  tab, she  seemed  to  do  well in that.

And  the  optimized  did  need   some  explanation,

I  think  just  because  that's a new word in her vocabulary.

But  I  think  that   she  had  that  sense  as  to,

okay,  these  were  the  settings that  were  going  to  be  the  best.

I  think  that  was  what  we  had to  talk  to  you  about  today.

I  did  just  have  a  few  final  questions   for  you,  if  that's  okay.

What  was  your  favorite  part   of  the  experiment?

Flying  the  airplane.

What  was  your  least  favorite  part?

Getting  hard  flying  the  airplane.

The air  was  about  100  degrees  Fahrenheit  when we were flying the airplanes,

but  we  didn't  have  a  lot  of  choice.

We  had  a  lot  of  storms coming  up  to  it,  didn't  we?

If  you  were  to  tell  somebody, what was  the  most  important  factor?

What's  the  most  important  thing if you  wanted  to  make  the  paper  airplane?

Maybe  play  type,  like,  with  a  dart.

Okay.

Was  there  anything  else that  was  important?

Nothing  you  can  really  think  of. Okay.

I  think  we  found   the  construction  paper a little bit,

and  I  think  it  did  actually  say   the  no  paper  clip.

I  think  you  were  saying   that  might  be  because  of  the  weight.

Bul I'll say yeah.

If  we  tried  it  again,

I think  you  had  the  paper  clip at  the  back  part  of  the  wing.

And  so  we  thought  about  maybe at the nose if  we  were  to  try  that  again.

But  let's  say,  what  do  you  want  to  do   for  your  next  experiment?

Should  we  do  another  one? What  would  you  want  to  do?

Statapult.

A  statapult? Yeah,  that  does  look  fun.

Ane now, I  didn't  ask  you to answer this in  any  way,  did  I?

But  was  Easy  DOE  easy  to  use?

Yes.

It was. Okay.

I'm  glad  to  hear.

I  think  that's  everything  that  we  have  left  today.

So  thank  you  for  your  time.

And  please  post  any  questions in  the  community  forums  below

if  you'd  like  to  ask either  of  us  anything.

Thank  you.

Comments
Phil_Kay

Love this, @Ryan_Lekivetz and Rory. It was a interesting to see how intuitive the analyze part was. I am a big fan of that interactive visual of the model.

mzwald

This is by far the cutest JMP webinar, well done!