Analisis Causa Raiz

Transparencias Tcnicas de Organizacin
Industrial: Metodologa para la seleccin y

gestin de proyectos. Fase III.
Rubn Rami Rivas
D-11/2008
Departamento de Organizacin de Empresas
Universidad Politcnica de Catalua
Publica: Edita:
Universitat Politcnica de Catalunya Ctedra Nissan-UPC

www.upc.edu www.nissanchair.com
admin@nissanchair.com
Fase 3. Anlisis Causa Raz
Departament
DOE dOrganitzaci
dEmpreses
TOI -10
Objetivo sesin
Definir objetivo y expectativas de la fase 3: Anlisis Causa Raz
Conocer herramientas ms usuales en la fase 3
Prctica con las herramientas
DOE www.nissanchair.com TOI -20

Fase 3: Anlisis Causa Principal
Identificar las causas principales del

Objetivo problema, y clarificar su relacin con
el 1er mtrico
Expectativas de esta fase:

Identificacin de las causas Causa Raz o Principal
Man A
B
Equipment
C
Material D
Method
E Impacto en el 1er Mtrico A B C D E
F
A
?
B
Target
Current value Root causes

Identificar la Causa Raz
EFECTO CAUSAS 1 CAUSAS 2
Ejemplo
1.Falta de validacin
DISE
DISEO fsica
DEFECTUOSO
2.Modificaciones de
ltima hora
NO SE 1.Falta de formacin
CONSIGUEN VALIDACI
VALIDACIN herramientas FV
OBJETIVOS DEFICIENTE
2.Falta prestaciones
CALIDAD
herramientas FV
1.Falta de validacin
fsica
ADIESTRAMIENTO
DEFICIENTE 2.Planificacin
deficiente
Establecer
Establecer la
la hiptesis
hiptesis desplegando
desplegando los
los efectos
efectos hacia
hacia las
las causas
causas principales
principales que
que los
los
provocan
provocan

Obtencin de datos para validar la hiptesis
# Mtodo Ejemplos
Rutina de gestin de
1 Datos obtenidos diariamente
datos
Asignando un observador
2 Datos obtenidos en el taller para evaluar un
determinado resultado
Entrevistas con clientes o conducir un Entrevistas sobre

3
cuestionario productos o servicios
Realizacin de una prueba, test o

4 Diseo de experimentos
simulacin
Es
Es crucial
crucial obtener
obtener datos
datos para
para validar
validar tanto
tanto las
las causas
causas como
como los
los efectos
efectos

Chequear la hiptesis
Puede ser
explicada Explicar lgicamente
lgicamente
Potencial
Causa Raz
Mtodo de
observacin Ej:
No es seguro
(Observar el efecto Es o No es
y/o causas
Usar mtodo Ej:

Estratificacin Pareto Diagrama
estadstico (basado
Histograma
en datos)
Ej:
Diagrama puntos
Correlacin entre Correlacin
causa y efecto Regresin
Anlisis de la Varianza
Determinar
Determinar una
una aproximacin
aproximacin cuantificable
cuantificable para
para especificar
especificar la
la causa
causa raz
raz

Determinar los factores que causan el
problema para especificar la causa raz
Proceso
Input Output Fase
Orientado a Fase 22
a b c d e f g h i j k Fase
Fase 33
Identificar
Identificar todas
todas las
las causas
causas
Ej:
A B C D E Brainstorming
Diagrama afinidad
Diagrama causa-efecto
Diagrama interrelaciones
rbol lgico
Especificar
Especificar causa
causa raz
raz
Ej anlisis cuantificable:
Pareto
Histograma
Regresin
A B C D E 55 65 75 85 95
Causas Raz Factor A
Cuantificar
Cuantificar siempre
siempre que
que sea
sea posible
posible los
los valores
valores cualitativos
cualitativos
Desarrollar la relacin entre la causa y el
efecto
Desarrollar la relacin entre el 1er mtrico y las causas principales con la intencin de estimar
el impacto y objetivo del 1er mtrico.
x = axA + xB + xC+
Clarificar
Clarificar relacin
relacin entre
entre 1er
1er
mtrico y causa raz
mtrico y causa raz
1st Mtrico Causas principales
Nivel de mejora
A
Causa A: 50%
Causa B: 30%
Causa C: 25% Estimar
B Estimar impacto
impacto
C Ej:. Anlisis cuantificable

Diagrama pareto
Grficos
Experiencia del equipo
Valor actual Valor Objetivo

Fase 3. Herramientas
ISHIKAWA
LOGIC TREE
DIAGRAMA INTERRELACION
QFD
Departament
DOE dOrganitzaci
dEmpreses
TOI -90
Ishikawa
Es la representacin organizada de la relacin
entre el EFECTO o problema y los posibles
DIAGRAMA
DIAGRAMACAUSA
CAUSA--EFECTO
EFECTO FACTORES que le afectan, estructurado en
forma de raspa de pescado.
Descripcin
ESPINA PEQUEA del problema
ESPINA
SUPERPEQUEA
ESPINA MEDIA
5 M,s 4M + Medicin
ESPINA GRANDE
Manode Obra
Materiales
4 M,s EFECTO
Mquinas
Mtodos
FACTORES (POSIBLES CAUSAS)

Ishikawa
Maquinaria/Equi Operario Atencin
Atencin a los
Mtodos de enseanza Deseo
Ajuste de utilaje detalles Agilidad Concentraci
Ejemplo de Diagrama de Causa y Efecto. Precisio
Ganas de trabajar
Habilidad
Aptitudes
Trabajo dificultoso
Desgaste Peridico Equipo de apoyo Empleado Paciencia
Utiles y
Principales reglas: herramientas
Atornillador
Programa
Al inicio
nuevo
Entrenamiento Experiencia
Habilidad
Fatiga
Capacidad de observacion
Dinamomtrica Habilidad para la

Mtodos de resolucin de
insoeccin Conocimiento Malos hbitos de vida
Recoger todos los factores posibles. Llave problemas

Puntos de inspeccin del producto Condiciones fsicas
Ubicacin Propietario de
Herramientas Procedimientos la tecnologa Enfermedad
Registro
Conciencia de
Capacidad de proceso
Los factores deben expresarse en trminos

Operacin Ambiente de trabajo
Tolerancia Confirmacin Mtodo
Conocimiento Ambiente familiar
QC
de la Calidad
Preocupaciones
concretos. Diseo
Habilidad
auditiva
Relacin con
Error en la
Capacidad de
operacin
Habilidad sensitiva Relacin con
trabajo los superiores los amigos
Los factores deben ser posibles causas, no

Capacidad de
mantenimiento Habilidad visual Daltonismo
Luces Ruido Polvo FIFO Proceso

Combinacin
contramedidas. Medio
Tiempo de espera
Organizacin
anticorrosin
Mtodos de
ambiente del trabajo
Durez Alargamiento control
Es muy conveniente clasificar los factores en

Primera pieza Contenedores
Temperatura Nuevo producto Ubicacin
Etiquetas
Organizacin Componente Volumen
Espacio disponible
las 4 M de la Produccin. rea de Cambio de Cambio de
del proceso
Dimensione Control de
stocks
trabajo de referencia Longitud Mtodo de inspeccin
Anchura
Layout Muestre
Las espinas de cada rama deben

Muestre Altur
Mtodo de Form
Balanc inspeccin Procesamiento de
defectos
mantenerse paralelas, para facilitar su Volumen de
trabajo
Auto-inspeccin Acabado
Rugosidad
Dobleces
Piezas
faltantes Operacin
Procedimient Puntos crticos
equivocada
visualizacin Volumen
Operacin estndar
Pieza
defectuosa
Defecto
Programa de
entregas Texto
Imgenes
Los factores enmarcados son
Mtodo de los considerados como
Mtodos escritura Materiales/piez particularmente importantes

Diagrama de Arbol
Es un diagrama que representa los E strategia 1.1.1

E strategia 1.1 O bjetivo
efectos y las diferentes causas E strategia 1.1.2
encadenadas que los producen, E strategia 1 O bjetivo

E strategia 1.2.1
ramificndose en forma de rbol. E strategia 1.2.2
O B JETIV O
E strategia 2.1.1
Bsicamente sirve para: FIN A L
E strategia 2.1.2
Clarificar las causas con E strategia 2 O bjetivo

E strategia 2.2.1
respecto a los efectos. E strategia 2.2 O bjetivo E strategia 2.2.2
Desarrollar estrategias y E strategia 2.2.3

(1er D esarrollo) (2 D esarrollo) (3er D esarrollo)
polticas para cumplir los
objetivos. F u n c io n e s p a r a
s a tis f a c e r la s 2 F a c to r 1 .1
p r im e ra s fu n c io n e s
F u n c io n e s p a ra 1 e r F a c to r 1
s a tis f a c e r la
Descomponer un objetivo F u n c i n d e
fu n c i n b s ic a
2 F u n c i n 1 .1 . C A L ID A D
R E Q U E R ID A 1
2 F a c to r 1 .2
p ro d u c to , e tc . 2 F a c t o r 2 .1
principal en sus elementos 1 F u n c i n 1
1 e r F a c to r 2
2 F a c t o r 2 .2
F U N C I N 2 F u n c i n 1 .2
bsicos. B S IC A
2 F a c to r 1
2 F u n c i n 1 .1 .
C A L ID A D
R E Q U E R ID A 2
1 F u n c i n 1 2 F a c to r 2
2 F u n c i n 1 .2

Diagrama de Arbol
Etapas para la elaboracin de un diagrama de rbol:
PASO 1: Elegir el tema
Describir el tema de forma clara y concreta, utilizando la
pregunta qu hay que hacer para reducir, mejorar,
eliminar,? y escribirlo en una etiqueta roja a la izquierda.
PASO 2: Indicar claramente las premisas Qu hay que
hacer para?
Identificar las premisas y escribirlas en etiquetas.
PASO 3: Identificar estrategias del primer nivel
De todas las estrategias de primer nivel, elegir de 2 a 4 y Estrategia de 1er
nivel
escribirlas en etiquetas. Tem a
Estrategia de 1er
PASO 4: Colocar las etiquetas en el papel nivel
Las estrategias de primer nivel se colocan a la derecha del

tema, en forma de rbol.

Diagrama de Arbol
Etapas para la elaboracin de un diagrama de rbol

(continuacin):
PASO 5: Identificar estrategias de segundo nivel

Las estrategias de primer nivel se convierten en
objetivos y se identifican las estrategias de segundo
nivel.
PASO 6: Desglosar todava ms las estrategias
De la misma forma que en el paso 5
PASO 7: Revisar la conexin de las estrategias
Tener en cuenta todos los niveles hasta llegar al tema
PASO 8: Ordenar las etiquetas y aadir
estrategias
PASO 9: Anotar datos como fecha, lugar, etc

Diagrama de Interrelacin
Efecto
Se utiliza para resolver problemas en Causa
Causa
Causa
los que intervienen causas y efectos

mezclados, en base a: Efecto Efecto
Causa Causa
Explorar todos los factores que Efecto Efecto
Causa Causa
puedan estar relacionados con el
problema y listarlos, indicando las
Efecto
posibles conexiones. Efecto
Causa
(Problema)
Confirmar las relaciones de causa
efecto entre los factores Causa
Efecto
Existen varios modelos:

Relacin entre causas y efectos.
Es conveniente aplicar esta
Relacin entre objetivos y herramienta en grupos de 6 a
estrategias / acciones para conseguir 10 personas.
los objetivos.

Etapas para la elaboracin de un diagrama de relaciones:
PASO 1: Elegir el tema (problema) TEMA:
Porqu.....?
Escribir el tema de forma clara y concreta, utilizando la pregunta por
qu? Y colocarlo en el centro de un papel grande en la pared.
PASO 2: Confirmar el tema
Confirmar entre todo el grupo antecedentes, premisas, etc.
PASO 3: Identificar los factores principales que causan el tema
(problema)
Identificar los factores principales y colocarlos dentro de un valo
alrededor del tema. Colocar flechas de causalidad de forma
provisional (con lpiz).
PASO 4: Desglosar las causas
Identificar las posibles causas de los factores mediante al anlisis
por qu?, por qu?, conectndolos con flechas provisionales.

Etapas para la elaboracin de un diagrama de relaciones

(continuacin):
PASO 5: Revisar las relaciones causales

Una vez establecidas las causas, revisarlas en conjunto y
confirmar las relaciones provisionales.
FACTORES IMPORTANTES
PASO 6: elegir los factores ms influyentes

Utilizar consenso o sistemas de votacin, por ejemplo, cada
asistente punta 5 de las causas con 3, 2 y 1 puntos y se
priorizan.
PASO 7: Escribir la conclusin en la parte superior
Las figuras representan los puntos
derecha
La conclusin debe ser consensuada ente el equipo y se
debe escribir de forma concisa. Indicar tambin los
miembros del equipo, la fecha, el lugar, etc.

Herramientas de soporte para realizar el diagrama de relaciones:

BRAINSTORMING
Para identificar los factores. Respetar lar reglas del Brainstorming:
No criticar
Aceptar positivamente las ideas libremente expresadas
Generar el mximo nmero de ideas
Comprender las ideas de los dems (escuchar)
Para relacin entre

SISTEMA DE LOS 5 POR QU? causas y efectos:
Causa
Efecto Factor 1 Factor 2 Factor 3 Factor 4
verdadera
Por qu? Por qu? Por qu? Por qu? Por qu?

QFD (Quality Function Deployment)
El mtodo QFD ha sido creado con el fin de traducir y dar

respuesta a los deseos o exigencias de los clientes, en parmetros
apropiados que sean entendibles por los tcnicos y empleados de
las empresas, todo bajo el prisma o paraguas que supone el nuevo
concepto de Gestin de Calidad, tal y como lo define el doctor K.
Ishikawa:
" La gestin de la Calidad consiste en desarrollar, concebir y

fabricar en los plazos previstos los productos y servicios ms
econmicos, ms tiles y ms satisfactorios para el consumidor."

Finalidad del QFD
1) La obtencin de una calidad de diseo excelente, mediante la

conversin de las necesidades del cliente en caractersticas de
calidad adecuadas, sin omisiones ni elementos superfluos.
2) El despliegue sistemtico de dichas caractersticas mediante

la bsqueda de unas funciones tcnicas:
- de los componentes del producto
- de los componentes del proceso
3) que estn lo mas correlacionadas posible con las necesidades del cliente.
Este es el lado tcnico del QFD. Para llegar a este punto, el QFD parte del
principio segn el cual las expectativas del cliente, todas pero "sin omisiones
ni elementos superfluos" son las que permiten programar el producto
adecuado en el momento adecuado.
Construccin del QFD
Correlacin entre los Comos
Productos/Servicios (Comos)
Ejemplo
Como espero conseguir las necesidades del cliente
Necesidades Anlisis
Anlisis de relaciones competencia
Cliente
Evaluar el nivel de satisfaccin de cada producto/servicio Omo nuestra
respecto a las necesidades de los clientes competencia satisface
Que es lo que espera (1 poca relacin, 3 media y 9 alta)
el cliente (QUE) las necesidades de los
clientes
Importancia de los productos
Especificacin de los
productos/servicios
Evaluacin tcnica
Historico

Ejercicio 8
Disea los factores principales
para poder conseguir un resultado
positivo en la prxima presentacin
a periodistas de nuestro nuevo
vehculo.
Utiliza la herramienta QFD, y
construye la casa de la Calidad 1er Departament
nivel) DOE dOrganitzaci

dEmpreses
Tiempo: 80 min (ejercicio) + 50 min

(todas las presentaciones)
TOI -220
Fase 3. Herramientas
F TEST
T TEST
ANOVA
CORRELACION
Departament
DOE dOrganitzaci
dEmpreses
TOI -230
Root cause analysis : Tools used
Brainstorming Process
Mapping
Affinity Diagram
Identify potential
root causes
5 Whys Fish Bone
Narrow down the Graphical Analysis

number of factors
Confirm statistically Statistical

the most important
factors Analysis
Statistical Analysis
issue1
issue1
issue2 issue3
Graphical Statistical
Brainstorming issue2 . issue4
issue3 . analysis analysis
/affinity
issue6
. issue10
issue 50
1 -Techniques such as brainstorming, process mappingetc, allow us to focus on areas of concern by sharing expertise among the
team members and therefore to prioritize the improvements according to the most important root causes.
2- Graphical analysis allows us to :

Either confirm some root causes : we can see it and not only we think it is an important root cause
Or focus the areas to brainstorm even before starting the analysis (for example, stratification).
The problem with graphical analysis is that it may be misleading (i.e take the wrong decision having read a graph) or not
explicit enough for people to make the good decision
3- Statistical analysis allows us to confirm or reject a hypothesis based on the most important causes, whilst
understanding that there is a risk of being wrong in the decision making process.
Because statistical analysis is time consuming and sometimes rather complex, it should be
done ONLY after pre selection of the potential vital few root causes.

Root cause analysis : Tools used
Statistical Analysis : Y = f(X)

1st metrics
1. Y is continuous, X is discrete:
T test to check the difference between the two
means; (X has two levels),
F test to check difference on variance,
Anova to test multiple means difference (X
has more than two levels) From ANOVA or
regression, calculate
2. Y is continuous, X is continuous: the weight of each
Correlation and regression (one X) factor (X)
Multi regressions (more than one continuous
X)
3. Y is discrete, X is discrete:
CHI2 (or proportions)
In
In each
each case,
case, use
use graphs,
graphs, test
test hypothesis,
hypothesis, test
test normality
normality and
and check
check conclusions
conclusions with
with experts
experts

Mean Test (also called t-Test)
What does it do?

When handling two sets of sample data, use this test to see if the
mean value of each population group is equal (or not).
When do you use it?

This test can be used to check if a factor has a real effect on the
mean during the root cause analysis phase.
It is also very useful to validate the impact of the improvement and
demonstrate the difference in the mean between prior and after the
change.
Limitation
As we are testing the mean, we need Normal distribution

One sample t-test
Population
N, m is known, but s is unknown
Sample :
n, x 1
What do we want to know ?

Is the sample pertinent for the
population from a mean
perspective?
Or, can we say that X1 = m ?
How can we check that ?

1. Check Normality of the sample
2. Compute the 1-sample t-test

1 sample t-test
How can we check that ? In MINITAB
Stats>Basic Statistics>Display Descriptive Statistics
Test of Normality
Summary for 30sample2

H0 = Sample is Normal A nderson-Darling N ormality Test
A -S quared 0,51
H1 = Sample is not Normal P -V alue 0,180
Risk max to be wrong = 5 % M ean

S tDev
318,45
8,21
V ariance 67,33
S kew ness -0,105891
Kurtosis -0,905940
N 30
Compute the test
M inimum 303,62
1st Q uartile 310,80
M edian 321,25
3rd Q uartile 323,23
305 310 315 320 325 330 335 M aximum 333,59
95% C onfidence Interv al for M ean
315,39 321,51
95% C onfidence Interv al for M edian
P-value < 5% P-value is P-value > 5 % 313,31 322,15
95% C onfidence Interv al for S tDev
calculated 95% Confidence Intervals
6,53 11,03
Mean
Median
312 314 316 318 320 322
Sample is not Normal Sample is Normal

1 sample t-test
How can we check that ?
1-sample t-test
H0 = mean is equal to 320

H1 = mean is not equal to 320
Risk max to be wrong = 5 %
Compute the test
P-value < 5% P-value is P-value > 5 %

calculated
Mean is NOT equal to 320 Mean is equal to 320

1 sample t-test: application
Boxplot of 30sample2
(with Ho and 95% t-confidence interval for the mean)
st!
lt fir
s u
re
i cal
a ph _
X
Gr Ho
300 305 310 315 320 325 330 335

30sample2
Confidence
interval on the
mean
X = Computed value of the mean HO: mean=320cm
Just looking at the graph, we can see that the 95% confidence interval overlaps 320cm so we
can already ensure we have a good chance that trial1 is relevant to the original population
from a mean perspective.

u lt In addition, Minitab gives the calculations on the session window
r es
al ! One-Sample T: 30sample2
c
ti nd
i s
t at eco Test of mu = 320 vs not = 320
S s
Variable N Mean StDev SE Mean 95% CI T P
30sample2 30 318,450 8,205 1,498 (315,386; 321,514) -1,03 0,309
Value of t (from the t distribution) Corresponding probability for 20 data points
Here we can say that we have 30% chance of being wrong by assuming that the
mean of 30sample2 differs from 320 and therefore, we will consider that the
calculated mean (318,45) is not different than the total population. Practically, we will
consider this sample of 30 data as a good representative of the total population.
If p-value (also called a) is less than 5%, we will say that the mean is different

2 sample t-test
We are now looking at 2 set of data and we want to check if they are identical in term of means
Sample 1 Before After Sample 2

Machine A Machine B
UK Spain
Dealer X Dealer W
Etc
n, x
1 1
n, x
2 2
Can we consider that X1 and X2 are identical?
x x 1 2
To answer this question we use a two sample t-test

Warning : this calculation is only
valid if each distribution is
normal
2-sample t-test
H0 = means are equal

H1 = mean are not equal
Compute the test

calculated
Means are different Means are equal

Total Lead time = f(Revisions)
st!
fir Boxplot of total leadtime by Revision
s ult 12
l re
ca
phi 10
a
Gr
total leadtime
median
6 Means
look
4 different
mean
2
0 1
Revision
Just looking at the graph it appears that the two means are different but
there is also quite an overlap: are they really different therefore?
Lets look at the statistical result to have the right answer

In addition, Minitab gives the calculations on the session window
ult
r es Two-sample
al ! T for total leadtime
c
ti nd Revision
i s N Mean StDev SE Mean
t at eco
S s 0 20 4,70 1,00 0,22
1 100 5,94 1,95 0,20
Difference = mu (0) - mu (1)

Estimate for difference: -1,24400
95% CI for difference: (-1,84081; -0,64719)
T-Test of difference = 0 (vs not =): T-Value = -4,18 P-Value = 0,000 DF = 52
Value of t (from the t distribution) Calculated P-value = actual risk

In that case, there is no risk (P-value=0) to say that the 2 means are
different
Redo the exercise with PIC of T-test is easy to use make sure your data are
revision and Total Lead time normal, and check the practical significance of your
result !

Variance Test (F-Test)
What does it do?

When handling two sets of sample data, use this test to see if the
variance of each population group is equal (or not).
When do you use it?

This test can be used to check if a factor as a real effect on the
Standard Deviation during the root cause analysis phase.
It is also very useful to validate the impact of the improvement and
demonstrate the reduction in process variation between prior and
after a change is implemented.
Limitation
As we are testing the Std Deviation, we need Normal distribution

F-test
We are now looking at 2 set of data and we want to check if they are identical in term of Std Dev.
Sample 1 Before After Sample 2

Machine A Machine B
UK Spain
Dealer X Dealer W
Etc
n , x ,
1 1, 1 n , x ,
2 2 2
Do my two samples have the same

standard deviation?
s1 s2
x 1 x2
To answer this question we use a F-test

F-Test: application
normal
F-test
H0 = Std Dev are equal

H1 = Std Dev are not equal
Compute the test

In this
example,
the
P-value < 5% P-value is P-value > 5 % samples
calculated are in two
different
columns
Std Dev are different Std Dev are equal

F-Test: application
and at
al ult
c
hi l res Test for Equal Variances for 30sample1; 30sample2
a p
Gr stica ce !
F-Test

Test Statistic 1,65
ati on
30sample1 P-Value 0,183
St Lev ene's Test

Test Statistic 1,65
P-Value 0,205
30sample2
Confidence
interval for each 6 8 10 12 14 16
95% Bonferroni Confidence Intervals for StDevs
case
30sample1
F test Calculation :
Because the p-value
is over 5%, we can 30sample2
conclude that there is

300 310 320 330 340 350
no difference of Data
variance between the

two samples Total Lead time = f(Revisions)

The basic principles of ANOVA (1/3)
We collected sales data from 3 different salesmen. The actual
margin is calculated and reported in the following chart.
Obtained
Obtained data
data and
and data
data plot
plot
Salesmen A1 A2 A3
A3
Margin 57 62 63 A2 Average
Yield of all
56 64 65 Y data
1 2 3 4
58 62 67 A1 1 2 3 4
57 60 65
Sum 1 2 3 4
228 248 260
0
Can we conclude from these

results that the salesman Or is it ordinary variation
influence the average of margin within several
output? orders ?
To answer this question we use the ANOVA test

Se3
(Variation within A3)
The
The Concept
Concept of
of Variation
Variation Decomposition
Decomposition
Y
Se2
(Variation
A3
within A2)
SA ST (Total
A2 (Variation
Se1 between) Variation)
(Variation
within A1)
A1
+ Se (variation within each

ST (Total Variation) = SA (variation between salesmen)
salesman)
What do we want to know ?
Do I see actual changes in the mean when I change the sales man ?
Is A1 really different from A2, from A3 and is A2 different from A3 ?
Is the local variation (with one sales man ) smaller than the variation seen when we change
the sales man ?
Is Se << SA ? How can we check that ? F-Test !!!

MINITAB
MINITAB will
will do
do all
all the
the calculation
calculation
S
STT (total
(total variation)
variation) =S
=SAA (variation
(variation of
of factor)
factor) ++ S
See (variation
(variation of
of
error)
error)
ANOVA
Source DF SS MS F P
Factor 2 130.67 65.33 32.67 0.000 <0.05

<0.05
Error 9 18.00 2.00
Total 11 148.67 F-Test !!!
These results indicate that the P value of the factor salesmen is less than 0.05. It can be
concluded that statistically speaking, the salesmen has a significant influence on the
margin.
Warning : this computation is valid only if each distribution is normal and if the
variances of each sample are equal
Application : One-way Layout ANOVA
We would like to know if the Delivery time is linked with the Transporter.
Question: has the factor Transporter a significant influence on the total
variability of Delivery time?
First lets check that the population is normal and then check that the variance of
the population is the same for each level of the factor Transporter : P-value??
Test for Equal Variances for Delivery time
Bartlett's Test
Akio
Test Statistic 6,88
P-Value 0,230
Levene's Test
Bunzo
Test Statistic 1,93
P-Value 0,095
Transporter
Ishiro
Kaoru
Kenjiro
Mazaaki
5,0 7,5 10,0 12,5 15,0 17,5 20,0

95% Bonferroni Confidence Intervals for StDevs

One-way Layout ANOVA
normal and has the same
variance
1 way ANOVA test
H0 = Means are equal = Factor is not significant

H1 = Means are not equal = Factor is significant
Compute the test

calculated
Factor is significant Factor is not significant

st!
fir
ul t Boxplot of Delivery time by Transporter
s
l re 50
ca
phi
a 40
Gr
Delivery time
30
20
10
0
Akio Bunzo Ishiro Kaoru Kenjiro Mazaaki
Transporter
From the box plot, we can already deduce that Kenjiro is in someway different
(from a mean stand point) from the other transporters.

l The factor transporter has 6 The Sum of the Squares represents the
ic a levels therefore 5 degrees of total distance between the mean at each
i st
t at ult !freedom level of the factor and the overall mean
S res d The famous n-1
o n
c The Mean Square is the variance
se between levels, it is calculated by
One-way ANOVA: Delivery time versus Transporter SS/DF (=2272/5)
Source DF SS MS F P
Transporter 5 2272,7 454,5 6,46 0,000
The p value gives the confidence
in the significance of the factor, in
Error 114 8015,3 70,3 this case, the overall variability is
Total 119 10288,0 truly influenced by the factor
transporter
There are 120 data points
therefore 119 degrees of The F is the ratio between the mean
freedom (and so only 119-5 square of the factor (454) divided by
= 114 DF for the error) the mean square of the error (70)
In this example, we are sure (p=0.0%) that the factor transporter is significant in the overall
observed variability, but this factor is not the only one as it explains only 2272/10288 = 22%.

Minitab also calculates other data
S = 8,385 R-Sq = 22,09% R-Sq(adj) = 18,67%
22.0% of the total variability is explained by the factor trial
Individual 95% CIs For Mean Based on Pooled StDev

Level N Mean StDev +---------+---------+---------+---------
Akio 20 36,000 8,020 (-----*-----) No overlap of the
Bunzo 20 32,300 7,672 (-----*-----) means between
Ishiro 20 30,850 6,976 (-----*------)
Kenjiro and Mazaaki
Kaoru 20 32,400 7,451 (-----*-----)
Kenjiro 20 21,800 11,542 (-----*------)
Mazaaki 20 29,650 7,829 (-----*------)
+---------+---------+---------+---------
18,0 24,0 30,0 36,0
Pooled StDev = 8,385
Conclusion: the factor transporter is significant in the overall variability observed in the
delivery time. Graphically we can see that Kenjiro is the one which creates the main difference.

Conclusion and traps : One-way Layout ANOVA
CONCLUSION :
ANOVA is one of the most popular tools for statistical root cause analysis :
Y= f(X).
It not only confirms the influence of the factor X in the overall variability (s) of Y
but it also indicates the weight of the factor.
FIVE COMMON TRAPS :

1. If the population is abnormal, calculations may lead to the wrong decision,
2. If the variances are not homogeneous, calculations may lead to the wrong
decision,
3. If each level of the factor does not contain at least 5 data points,
calculations may lead to the wrong decision,
4. If the factor is continuous (such as pressure, time..) and not discrete,
calculations may lead to wrong decision.
5. If the factor looks graphically significant but the p- value is not good (for
example 20%), it is may be due to the sample size being too small; in that
case, add more data and re-check.

3.6) Correlation Analysis and Regression
Analysis
1. What Does It Do?
Scatter diagrams are used to describe the relationship of two

quantitative variables. Correlation analysis is conducted to
measure the correlation of the two variables.
Regression analysis is conducted if there is correlation between
the two quantitative variables. Regression analysis determines the
equation of the two variables linear relationship.
Correlation analysis: Measures

the strength of correlation.
Regression analysis: Draws a
straight line in the scatter
diagram.
Correlation Analysis and Regression
Analysis
2. Where Do You Use It?
Investigate whether there is a correlation between the variables

with correlation analysis. Then investigate further for cause-and-
effect relationships.
Use regression analysis where there is a correlation between two
variables. See the following Examples.
When you want to control output variables with input variables

Correlation Coefficient
Correlation coefficient: R is a statistic that describes the

strength of a linear relationship between two data.
Correlation coefficient: r can take values between 1 and +1. Correlation gets
stronger as the absolute value of R gets closer to 1.
Also, Correlation coefficient: R>0 indicates positive correlation
Correlation coefficient: R<0 indicates negative correlation
Correlation coefficient: R=0 indicates no correlation
Guidelines for Correlation coefficient: R
-1 -0.8 -0.2 0 0.2 0.8 1
Strong Negative Moderate Negative Weak or no Moderate Positive Strong Positive

Correlation Correlation Correlation Correlation Correlation

Scatter Diagram
Correlation coefficient and scatter diagrams are shown below
No
No correlation
correlation Moderate
Moderate positive
positive correlation
correlation Strong
Strong positive
correlation
Correlation
Correlation coefficient=0
coefficient=0 Correlation coefficient=0.6
Correlation coefficient=0.6 Correlation coefficient=0.8
Correlation coefficient=0.8
Perfect
Perfect positive
correlation Strong
Strong negative
negative correlation
correlation Non-linear
Non-linear correlation
correlation
Correlation coefficient=1.0
Correlation coefficient=1.0 Correlation coefficient=-0.8
Correlation coefficient=-0.8 Correlation coefficient=0
Correlation coefficient=0

Analysis
Regression analysis
H1 = Regression is significant
H0= Regression is not significant
Compute the test

calculated
Regression is significant Regression is not significant

Analysis
Fitted Line Plot

Transportation Cost = 156,3 + 0,6861 Volume
S 2,40378
185 R-Sq 84,6%
R-Sq(adj) 83,9%
a l 180
c
hi st !
Transportation Cost
a p ir
Gr ult f 175
s
re
170
165
160
0 10 20 30 40
Volume
We can see a trend and the relationship is very strong confirmed by a r factor of 83.9 % (adjusted for
sample size)

Analysis
ical !
st ond
Regression Analysis: Transportation Cost versus Volume
at i c
St t se
The regression equation is
sul Transportation Cost = 156,3 + 0,6861 Volume
r e
S = 2,40378 R-Sq = 84,6% R-Sq(adj) = 83,9%
Analysis of Variance
Source DF SS MS F P
Regression 1 698,714 698,714 120,92 0,000
Error 22 127,120 5,778
Total 23 825,833
P-value is less The R factor

than 5%, then the indicates the
regression is strength of the
significant relationship between
weight and size
Analysis
Fitted Line Plot

Transportation Cost = 156,3 + 0,6861 Volume
190 Regression
95% C I
95% PI
S 2,40378
180 R-Sq 84,6%
Transportation Cost
R-Sq(adj) 83,9%
170
160
Option: display confidence intervals
Allows to better assess the confidence on
150
the regression line and for individual data 0 10 20 30 40
Volume
point.
For example, if the volume is 20, there is
95% chance to have a transportation cost
between
165 and 175.
Use of confidence intervals

Analysis
Residual Plots for Transportation Cost

Normal Probability Plot of the Residuals Residuals Versus the Fitted Values
99 5,0
90 2,5
Residual
Percent
50 0,0
10 -2,5
1 -5,0
-5,0 -2,5 0,0 2,5 5,0 160 170 180
Residual Fitted Value
Histogram of the Residuals Residuals Versus the Order of the Data

5,0
4
3 2,5
Frequency
Residual
To display residual analysis: this allows to 2 0,0
check for special pattern which may lead to 1 -2,5
0 -5,0
wrong interpretation of the correlation or to -4 -2 0
Residual
2 4 2 4 6 8 10 12 14 16 18 20 22 24
Observation Order
suspect hidden factor.
Use of residual analysis

Tips for Conducting Correlation Analysis and
Regression Analysis (1/2)
(1)Examine
(1)Examine the
the data
data pictorially
pictorially with
with scatter
scatter diagrams
diagrams before
before correlation
correlation analysis
analysis and
and
regression
regression analysis.
analysis. The
The points
points to
to check
check when
when you
you are
are examining
examining scatter
scatter diagrams
diagrams
are:
are:
Is
Is there
there any
any need
need for
for stratification?
stratification? Diagrams
Diagrams a,
a, bb
Are
Are there any non-linear relationships?
there any non-linear relationships? Diagram
Diagram dd
Are
Are there
there any
anya)dislocated
dislocated points?
points? Diagrams
Diagrams
b) e,
e, ff
Always
Always draw
draw scatter
scatter diagrams.
diagrams. There
There could
could be
be several
several different
different cases
cases with
with the
the same
same
correlation
correlation coefficient
coefficient
c) and
and regression
regression equation,
equation,
d) shown
shown below.
below.
14 14
12 y = 0.5x + 3.0 12 y = 0.5x + 3.0
10 r = 0.82 10 r = 0.82
8 8
6 6
4 4
2
f)
e) 2 4 6 8 10 12 14
2
2 4 6 8 10 12 14
14 14
12 y = 0.5x + 3.0 12 y = 0.5x + 3.0
10 r = 0.82 10
r = 0.82
8 8
6 6
4 4
2 2
2 4 6 8 10 12 14 2 7 12 17 22

Tips for Conducting Correlation Analysis and
Regression Analysis (2/2)
Common pitfalls
(1)
(1) Skipping
Skipping scatter
scatter diagrams
diagrams and
and going
going onto
onto correlation
correlation analysis
analysis and
and
regression
regression analysis.
analysis.
(2)
(2) Jumping
Jumping toto the
the conclusion
conclusion that
that there
there is
is no
no correlation,
correlation, because
because the
the
correlation
correlation efficient
efficient is
is small.
small.

There
There are
are cases
cases in in which
which there
there seems
seems to
to be
be no
no or
or weak
weak correlation,
correlation, when
when in
in
fact
fact there
there is
is aa strong
strong one.
one. This
This happens
happens when
when the
the analysis
analysis scope
scope is
is too
too small.
small.
(3)
(3) Jumping
Jumping to
to the
the conclusion
conclusion that
that there
there is
is aa cause-andeffect
cause-andeffect relationship,
relationship,
because
because there
there is
is aa correlation.
correlation.

A
A strong
strong correlation
correlation between
between two
two variables
variables does
does not
not necessarily
necessarily mean
mean that
that
there
there is
is aa cause-and-effect
cause-and-effect relationship.
relationship. Judging
Judging whether
whether there
there is
is aa cause-
cause-
and-effect
and-effect relationship
relationship requires
requires additional
additional analysis.
analysis.
(4)
(4) Extending
Extending the
the regression
regression line
line and
and predicting
predicting values
values outside
outside the
the scope
scope
(extrapolation).
(extrapolation).

There
There is
is no
no guarantee
guarantee that
that the
the regression
regression line
line extends
extends beyond
beyond the
the data
data
obtained.
obtained. You
You will
will need
need technical
technical examination
examination for
for extrapolation.
extrapolation.

Ejercicio 9
Anlisis Grfico: Analiza la
relacin entre descuento y margen
entre las regiones y productos.
Justifica las conclusiones
Departament
(Fichero: anlisis grfico.xls)
DOE dOrganitzaci
dEmpreses

TOI -610
Ejercicio 10
t & F Test: Analiza la resultados
de los costes de garanta de dos
pases
Departament
DOE dOrganitzaci
dEmpreses
(Fichero:t y F Test.xls)

TOI -620
Ejercicio 11
ANOVA: Evala el impacto en la
media de diferentes modificaciones
realizadas sobre una mquina
Departament
DOE dOrganitzaci
dEmpreses
(Fichero: ANOVA.xls)

TOI -630
Ejercicio 12
Correlacin y regresin: Explica
la relaciones entre coste y volumen
del transporte determinado
Departament
DOE dOrganitzaci
dEmpreses
(Fichero: correlacin y regresin.xls)

TOI -640
Ejercicio 13
Analiza los datos de carga las
bateras, y comenta si justifican o
no la causa raz
(Fichero: Pepe car baterias.xls)

Departament
DOE dOrganitzaci
dEmpreses

TOI -650
Ejercicio 14
Justifica cual es la causa o
causas principales de la falta de
precisin en el plan de ventas de la
compaa
Departament
(Fichero: precisin ventas.xls)
DOE dOrganitzaci
dEmpreses

TOI -660

Analisis Causa Raiz

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Analisis Causa Raiz

Enviado por

Direitos autorais:

Formatos disponíveis

Transparencias Tcnicas de Organizacin

Industrial: Metodologa para la seleccin y

Rubn Rami Rivas

Departamento de Organizacin de Empresas

Universidad Politcnica de Catalua

Universitat Politcnica de Catalunya Ctedra Nissan-UPC

Definir objetivo y expectativas de la fase 3: Anlisis Causa Raz

Conocer herramientas ms usuales en la fase 3

Prctica con las herramientas

DOE www.nissanchair.com TOI -20

Identificar las causas principales del

Expectativas de esta fase:

Current value Root causes

DOE www.nissanchair.com TOI -30

DOE www.nissanchair.com TOI -40

Entrevistas con clientes o conducir un Entrevistas sobre

Realizacin de una prueba, test o

DOE www.nissanchair.com TOI -50

Usar mtodo Ej:

DOE www.nissanchair.com TOI -60

1st Mtrico Causas principales

C Ej:. Anlisis cuantificable

DOE www.nissanchair.com TOI -80

FACTORES (POSIBLES CAUSAS)

DOE www.nissanchair.com TOI -100

Dinamomtrica Habilidad para la

Recoger todos los factores posibles. Llave problemas

Los factores deben expresarse en trminos

Los factores deben ser posibles causas, no

Luces Ruido Polvo FIFO Proceso

Es muy conveniente clasificar los factores en

Las espinas de cada rama deben

DOE www.nissanchair.com TOI -110

Es un diagrama que representa los E strategia 1.1.1

efectos y las diferentes causas E strategia 1.1.2

encadenadas que los producen, E strategia 1 O bjetivo

Clarificar las causas con E strategia 2 O bjetivo

respecto a los efectos. E strategia 2.2 O bjetivo E strategia 2.2.2

Desarrollar estrategias y E strategia 2.2.3

DOE www.nissanchair.com TOI -120

Las estrategias de primer nivel se colocan a la derecha del

DOE www.nissanchair.com TOI -130

Etapas para la elaboracin de un diagrama de rbol

PASO 5: Identificar estrategias de segundo nivel

DOE www.nissanchair.com TOI -140

los que intervienen causas y efectos

Existen varios modelos:

DOE www.nissanchair.com TOI -150

DOE www.nissanchair.com TOI -160

Etapas para la elaboracin de un diagrama de relaciones

PASO 5: Revisar las relaciones causales

PASO 6: elegir los factores ms influyentes

DOE www.nissanchair.com TOI -170

Herramientas de soporte para realizar el diagrama de relaciones:

Para relacin entre

DOE www.nissanchair.com TOI -180

El mtodo QFD ha sido creado con el fin de traducir y dar

" La gestin de la Calidad consiste en desarrollar, concebir y

DOE www.nissanchair.com TOI -190

1) La obtencin de una calidad de diseo excelente, mediante la

2) El despliegue sistemtico de dichas caractersticas mediante

Correlacin entre los Comos

DOE www.nissanchair.com TOI -210

nivel) DOE dOrganitzaci