SPSS Notes

Open Files \ Import Data from

Excel:

Steps:

 File  open  data

Terms in Hebrew:

 ‫ ממוצע‬mean
 ‫ חציון‬,median
 ‫ שכיח‬mode
 ‫ וסטיית התקן‬Standard Deviation
 ‫ אחוזים‬percentage
 ‫ ( ציון תקן‬standardized values – Z score
mark V in the box at the bottom of
Descriptives – look up)
 ‫ דרגות חופש‬degrees of freedom
 will be the number of groups
less than 1
 ‫ – מסולם רווח\מנה‬scale data
Data is presented to two decimals with rounding 5 up   ‫ – שמי‬nominal data
3.06 = 3.1  ‫ סדר‬- ordinal data
 ‫ שונויות‬- variance
Steps in Preparing Data for Analysis:

 Recode: Re-encoding of opposite questions (or different metrics)

o Transform  Recode into different variable
o Used for revering scores in likert scales
o Be sure to re-label in the variables window, make ordinal, and set decimals to 0

 Count Variable: Creates a variable that tells you how many times certain conditions
(such as a number or name) appear in a selected variable
o Transform  Count values within cases
o Counts number of times a response shows up (times person answer “never”)

 Missing variables

If you do not want to count cases with missing vairables go into compute and write
MEAN.X() X means the number of required entries to be counted within selected data
  example: MEAN.4(S1,S2,S3)

 Label

recode
 Compute: Create a new variable from a number of raw
variables
o Create an average / sum or any other manipulation of
variable \ variables
o Transform  Compute variable function group all
 function and special variables mean  replace ?
with the variables you want to create an average with
o Used for creating an average for a group of like scales
(transforming 5 different likert scales assessing
satisfaction to create one general satisfaction score)

Hypothesis Testing
Hypothesis:

 H0: "null hypothesis": the existing situation until proven otherwise.

o There is no connection between variables
 H1: "Research hypothesis": what the researcher wants to examine. That is, H0 is incorrect.

Errors:

 Type 1 error (alpha error): I rejected the null hypothesis although it was correct
 Type 2 error: I did not reject the null hypothesis, although it is incorrect

Alpha – levels of significance (‫)רמת מובהקות‬

 The probability of a wrong decision when H0 is correct

o In the social sciences, it is customary to use a 5%
o When less than 5% chance of error - finding is significant and accept H1

Types of hypothesis:
  Two-tailed hypothesis: A connection hypothesis without direction: more stringent
 Single-tail hypothesis: Conjecture contact with direction: less stringent

When reporting results:

 Alpha (p <0.05) means the results reject the H0 in favor of the Research Hypothesis
 Freedom degrees (‫)דרגות חופש‬
 Effect size (‫)גודל אפקט‬

Parametric and non-parametric tests:

 Parametric test – scale data

o robust data such as averages and variances
o A normal distribution (in this case, in each of the groups)
o Random sampling
 Non-parametric – nominal and ordinal data
o fewer demands on population – doesn’t assume population is distributed normally
o Chai Squared Test

Chai Squared Test

1) examines relationship between two nominal variables or

one nominal and an ordinal scale
2) examine the distribution of one variable
a. ex: Probability of a child daughter or son,
percentage of right, left and center voters

 Always two-tailed hypothesis ( ex: Is there a relationship between location and anxiety)
o Case Example: is there a connection between eye color and preferred learning style.
 Participants: 100 participants
 Independent variable: Eye color (brown, blue or green)
 Dependent variable: Preferred learning style (face-to-face, online or both)
 Ho: no connection between eye color and preferred learning style.
 H1: is a link between eye color and preferred learning style.
o Requirements:
 Random sampling
 Independence between observations
 Foreign and exhaustive categories
 No more than 20% of cells with a predicted incidence are less than 5
 does not indicate the intensity of the effect size but only if significance exists
 When the difference between the expected value and the actual value is less than 5, this
difference is not significant (20%)
Steps

Analyze  Descriptive statistics  Crosstabs

Row(s) = Independent variable Column(s) = Dependent

variable

Statistics Window:

 Chi square
 Nominal = Phi and Cramer’s V

Cells Window:

 Counts = Observed, Expected

 Percentages = Row

Tables:

Compare count with expected count

How to write equation:

 χ2 (4, N =100) = 13.00, p = 0.01

 df = 4 { (r-1) * (c -1) = 4}
 Establishing the Effect Size in Chai
Squared:

 phi test only suitable for

cases of 2X2 - always use
Cramer's v because it’s
suitable for all cases.

Note number in the values column for effect

size

How to Write Data Outcome:

A two factor analysis was conducted to examine if there’s a link between eye color and preferred
learning style. The two variables were eye color (brown, blue or green) and preferred learning style (face
to face, online or both).

The two variables were significantly related, 22 (4, N = 100) = 13.00, p = 0.011.

findings showed face-to-face learning style was most preferred by those with brown eyes (41.9%), blue
eyes (33.3%), and a slight preference for green eyes (6.7%).

The highest preference for online style was among those with green eyes (50.0%), but other colors also
had high percentages: brown eyes (45.2%) and blue (41.0%).
The highest preference for both learning styles was found among those with green eyes (43.3%) while
the numbers were lower in blue eyes (25.6%) and walls (12.9%).

The intensity of the association between eye color and preferred learning style was high, as measured
by the Cramer's V = 0.25.

Second Use of this Test: examine the distribution of one variable (‫)טיב התאמה‬

Case: whether there is a connection between color distribution of eyes in the sample of the research
and if it was generalizable to the general population.

 Participants: 100 participants

 Independent variable: none
 Dependent Variable: Eye color (brown, blue or green) (categorical or ordinal)
 Relative to: Percentage of eye color in the population: (82 = brown, 13 = blue, 5 = green)

Steps:

Analyze  Nonparametric tests  Legacy dialogs  Chi-Squared…

How to Write Findings:

A single-factor analysis was performed to examine if the size of the groups of eye colors in the sample
matched the distribution of eye color in the population.

A significant discrepancy was found between the size of the groups in the sample and the expected
population: 22 (2, N = 100) = 208.72, p <0.001.

findings showed there was a much lower sample of brown eyes (31%) than expected in the population
(82%). This reversal was found in blue and green eye colors (39 and 30 percent) in which the predicted
was 13 and 5 percent respectively.

The intensity of the effect found (ie, the size of the sample's non-conformity to the population) was very
strong, as measured by the Kramer index for examining the effect size: Cramer's V = 1.02
 T-Tests

 parametric test
 compares data to normal distribution
 Is based on:
o The gap between the two averages
o Variance of the variable
o And sample size


Indices of normal distribution

1) The Skewness Index - How symmetrical/asymetrical the distribution is - positive or negative

2) Measure of Kurtosis - flatness or steepness of peek of graph

Steps:

 DATA-T  Analyze  Descriptive statistics Descriptive

 Options-> Skewness & Kurtosis
 Variable: hourtot
 One Sample T-Test

One Sample T-Test: compares data obtained in sample to a known average

 similar to chai squared test except the variables are higher quality (scale)

Case Example:

researcher collected data on weekly hours of television viewing of a representative sample of 60

Jerusalemites, and compared to average weekly hours of the country collected in another study (M =
22.00).

Null Hypothesis can be two tailed or one tailed or

Steps:

 Analyze  Compare Means  One-Sample T-Test

 Test variable: hourtot
 Test value: 22
 x
t
s
n (Report: N, Mean, SD)

(t, df (df=N-1), Sig (two-tailed) test value)

Effect Size: Cohen’s D:


 Effect size table (ignoring positive / negative sign):
o Up to 0.2: negligible
o 0.2-0.5: Weak
o 0.5-0.8: Medium
o 0.8 or higher: strong

How to write findings:

A one-sample t-test was conducted to assess whether the average number of hours of television viewing
among Jerusalem residents was 22, the general average of Israeli citizens.

The findings show that the average Jerusalem sample (M = 22.78, SD = 2.71)

Significantly higher than the 22 (the average known to us in the rest of the country);

t (59) = 2.24, p = 0.029. The effect size (d = 0.29) indicates a weak effect.

Paired-Samples t-test

Paired-Samples T-Test: used when the observations are paired; compare two measurements that
belong to the same unit

Example:

 difference between the averages of the same people in two different courses
 difference in the ratio of students to distressed families before and after a year of practical work
o T1 to T2
 Husband and wife or parent and child comparisons
o Group Comparison

Hypothesis:

 (H1): Young people watch more hours on average on Thursdays than on Sundays
 (H0) Thy Watch the same amount of hours/no difference
 o one-tailed hypothesis

(Mean, N, SD for both)

(t, df, sig 2-tailed)

Effect size (Sig. 2-tailed) (ignoring positive / negative sign) (Same as One-Sample T-Test):

 Up to 0.2: negligible
 0.2-0.5: Weak
 0.5-0.8: Medium
 0.8 or higher: strong

How to Write Findings:

A paired samples t-test was conducted to assess whether the television viewing of young people on
Thursdays is higher than the television viewing on Sundays. The results indicate that the average level of
viewing on Thursdays (M = 3.78, SD = 1.12) is significantly higher than the average TV viewing on
Sundays (M = 3.32, SD = 1.29); t (59) = -2.29, p = 0.013. The effect size (d = 0.30) indicates a weak effect.

Independent Sample T-Test

Independent Sample T-Test - When we want to compare the averages of two independent groups

 Independent Variable: Dichotomous nominals like gender, or parents/children

Examples:

 Comparison between the experimental group and the control group

 Compare kids with regular parents and kids with hearing loss parents, and their social abilities

Hypothesis:

 (H1): is a difference between women and men in number of times they fall asleep in front of TV
 (H0) no difference between women and men in number of times they fall asleep in front of TV
 Independent Variable – Gender
  Dependent Variable – number of times falling asleep

Steps:

 Analyze  Compare Means  Independent-Samples T-Test

 Test Variables = sleep
 Grouping variable = sex
o Group 1: 1 (men)
o Group 2: 2 (women)

N,M,SD

Levin’s Test

 designed to examine the hypothesis of equality of variance (‫ )שוויון השונויות‬Are the two groups
we are comparing actually comparable – is the data roughly distributed in the same way or not)
 Levene's Hypotheses
o H0 the equality of variance is assumed p>.05
o H1 the equality of variance is not assumed p<.05
 If results of Levin test are significant, - there is no equality of variance, so we will use the second
row of output in which there is a correction to the test, which allows us to report the results
despite the violation of equality. If it is not significant, stick with the upper row

sig, t,df,sig 2-tailed

Effect Size:

 slightly different because it takes into account the size of the two groups
 d = t*√[(N1+N2)/(N1*N2)] = -13.143 / [(31+29)/(31*29) = -3.40
 Effect Size Scale:
o ‫ זניח‬:0.2 ‫עד‬
o ‫ חלש‬:0.2-0.5
o ‫ בינוני‬:0.5-0.8
o ‫ חזק‬:‫ ומעלה‬0.8

How to write findings:

An independent t-test was conducted to test the hypothesis that there is a difference between
men and women in the number of times they fall asleep in front of the TV per month.

The test showed a significant difference between men and women at their level of recurrence (t
(58) = -13.14, p <0.001). Men reported an average of less than (M = 5.68, SD = 1.25) compared
to women (M = 13.69, SD = 3.14). The effect size found (d = -3.40) indicated a very strong effect.

ANOVA – Univariate One Way ANOVA

Assumptions in the test:

 Independent Variable – three different nominal categories

 Dependent Variable – Scale
 Normal distribution of the dependent variable in each group
 The assumption of equality of variances (we will examine with the Levin test pessimist)
 Independence between observations
 Random sampling

Examples of the F Test:

 any differences between the types of ownership of a welfare organization (government,

business, nonprofit) at the level of customer satisfaction?
 difference between different types of therapists (social worker, psychologist, art therapist) in
the effectiveness of treatment?
 differences between different intervention to reduce anxiety (cognitive therapy, exposure
treatment, online therapy, drug therapy) in reducing anxiety levels from statistics?

ANOVA is looking at the differences within the group and between groups

 greater the variance between groups (explained variance) than the variance within the groups
(unexplained variance - the variance of the error): the evidence for real (non-random) errors
between the groups is "more convincing".
 Another formulation: a large difference between sample averages and smaller within the scores
of each group: leads to greater confidence in probabilistic conclusions (and a smaller probability
of error in obtaining 1H).

How the test actually functions:

 The test will examine the relationship between the explained variance and the error variance
 The effect size will be the ratio of the explained variance to the general variance
 The degree of freedom is required for the test:
o dfwithin (error) = ‫( סך התצפיות‬n) – ‫( מספר הקבוצות‬k) ‫בתוך‬
o dfbetween = ‫( מספר הקבוצות‬k) – 1 ‫בין‬
o dfTotal = ‫( סך התצפיות‬n) – 1 ‫כולל‬
Definitions:

 Variance between groups

o MSB (Mean Sum of Squares Between)
o (between = explained): The difference between groups is a difference explained by the
values of the independent variable.
 The variance within groups:
o MSW (Mean Sum of Squares Within)
o (within = unexplained; Error): The difference within groups is an unexplained difference,
we do not know what the difference is. Sometimes it is called a different error or MSE
 The general variance of the research system:
o MST (Mean Sum of Squares Total)
o Value of total difference = within (unexplained difference) + between (explained
difference)

Case Example:

 examine whether type of treatment child was receiving had an effect on his mental well-being.
o Participants: 30 children
o Type of treatment:
 Without treatment (control group)
 Short treatment
 Long treatment
o Dependent variable: the degree of mental well-being a year later
 (An 8-point scale, 1: low, 8 high)

Hypothesis:

 always two tailed

 In our example:
o H0 no connection between type of treatment and level of mental well-being
o H1 is a connection between type of treatment and level of mental well-being

Steps:

Analyze  General Linear Model (GLM) 

Univariate

Fixed factor – independent variables with nominal

values
 Options
 Descriptive Statistics
 Estimates of effect size
 Homogeneity tests

If Sig < .05 (Levine Test is significant) the variances are not equal so you run a Dunnet T

If Sig > .05 (Levine Test is not significant) the variances are equal so you run a Tukey

if Sig> .05, the assumption of variance being equal is preserved, and we continue the test. In
Anova, even if assumption is not maintained, we continue the test but will have to change
something in the analysis, similar to what we did in the T test. (Tukey)
 look at Tipil and error

 row of the independent variable (Tipul)

o Describes the differences between the groups
 Error row (always appears)
o Describes the data within the groups themselves

Between groups = K-1 legend: K Number of groups, N Number of participants

Within the groups = N-K

Effect Size:

 The Eta value assesses effect size

 Going from 0 (lack of correlation) to 1 (full correlation) when:
o Between 0.01 and 0.05 - small effect
o Between 0.06 to 0.13 - medium effect
o 0.14 and above - large effect

IMPORTANT: Unlike Cohen's D, in this case the number has significance: the independent variable
explains h2 percent of the variance in the dependent variable (this is why it cannot go through 1 and
note how this is reflected in the calculation of the index)

F-Test

 F not significant - only report F, degrees of freedom, and significance

  F significant - report an F value, degrees of freedom and significance
o Then report - effect size, post hoc tests (Tukey or other, as the context may be)

Post-Hoc:

 Unlike T test, test results do not describe where the

difference is. When the F test is significant, perform a test
to find out where the significant differences are.
 Post Hoc Tests for:
o ‫ בדוגמא שלנו( המשתנה הבל"ת‬Tipul)
 Levine not significant =Equal Variances
Assumed – Tukey
 Levine significant = Equal variance not
assumed - Dunnett’s T3

The null hypothesis (H0) was rejected: at least one group was different from the other groups

An asterisk (*) in the mean difference column (Mean Difference) means a significant difference in the
level (significance of .05 is also expressed in the sig column

Compare no tipul vs. tipul katzar

 Another way (easier) to look at post hoc analysis (appears only in
TUKEY)

 numbers appear in same column: there is no significant

difference between the groups
 numbers appear in different columns: there is a
significant difference between the groups

How to write findings:

Univariate One-Way ANOVA was conducted to examine relationship between type of treatment child
was given and level of mental well-being.

independent variable included three types of treatments: no treatment (control group), short-term
therapy and long-term treatment. The dependent variable was the level of mental well-being of the
child at the end of the year.

The result of the analysis of variance is significant (F, 2, 27) = 40.50, p <.001. The intensity of the effect,
as assessed by a square, was found to be high, with the type of treatment explaining 75% of the variance
in the child's mental well-being

To examine the differences between the different levels of the independent variable, we chose to make
post-hoc comparisons using the Tukey index (after a preliminary test that did not show a significant
difference between variances).

It appears that the average mental well-being of children in long-term care is significantly higher (M =
6.60, SD = 1.07) than children in short-term care (M = 3.60, SD = 0.84) and no treatment (M = 3.00, SD =
0.94). No significant difference was found between the last two frames.

long-term therapy is more beneficial for mental well-being than the other two types of treatment.

Working with subgroups

 Work on part of the database, according to different criteria:

o If
o Filter variable
o And more ...
 Do not forget to always cancel at the end of the selection (may appear in the test)

Split cases:
 o Data  Select cases  condition is satisfied If (select case you want to isolate and
type like so: Xvariable=1 (if 1=bachlors, 2=masters, and 3=doctorate) then run bivariate
correlation

Split file is similar to select cases but analyzes the different groups at the same time and they will be
displayed separately in the output.

 USE: When you want to isolate nominal data to run correlation on each separately
o Ex: education level (bachelors, masters, doctorate) to 3 different verbal tests
 two options for splitting, an option for comparison analysis and a separate analysis option
(Organize output by groups). The outputs are similar, use convenience.
 Try to split the file into the sample, and see if there are changes in the pattern of the
relationship in your analysis.

Split file

Split file comparison analysis or separate analysis option  select the variable you want to separate
output on (education level)  bivariate correlation

Correlations

stronger tests designed for scale variables

 When there is at least one scale variable, we can look for cases in which a change in the value of
one variable results in a change in the value of another variable - correlation
 There is no dependent variable and independent variable, only a relationship (symmetric index).

For example:

 Place in the competition and budget invested

 height and weight
 Height and IQ
 Number of errors in the test and score

Pearson and Spearman measure direction and intensity of a linear relationship between two variable
when assessing a scale variable.

 Pearson: both variables are scale

o cannot be made on a variable that does not have an average
o assesses if correlation is linear
o examples of questions for this measure:
 level of education in years of schooling and the level of monthly income
 age of child and the number of cases of violence to which he participated
 Number of hours of study per exam and test score
 Spearman: When both variables are
o Order
 o One is scale and the other is order

The stronger the relationship between the variables, the more accurate the prediction.

Need to assess:

 Relationship type
 Context direction: direct (positive), reverse (negative)
 The strength of the connection (no connection, weak link, strong bond)
 Context (linear, non-linear)

Example of cases we can look for correlations:

 The level of education in the number of years of schooling and the level of monthly income
 The age of the child and the number of cases of violence to which he participated
 Number of hours of study per exam and test score
 Number of cases of violence against social workers and satisfaction with work

1 and -1, means full linear connection, that is, there is no error

farther away from 0 and closer to 1 or -1 - stronger the relationship

becomes, and the smaller the error

 0 - 0.1 - negligible / trivial

 0.1- 0.3 - weak
 0.3 - 0.5 - medium
 0.5 and higher strong
the correlation is also the effect size!

The statistics (and the size of the effect) indicate both the intensity of the relationship and the direction
of the relationship (positive or negative)

The relationship is linear. There may be a non-linear connection and then no adapter is found or no high
correlation is found

Steps:
Analyze  Correlate 
Bivariate  select
correlation test  options
means and SD

Missing values

 Pairwise option
 Listwise option

 Pairwise: If there are more than two variables and want as many observations as possible
(comparing gender to height and weight but participant 6 didn’t fill in weight. SPSS will still
compare gender and height for that participant)
 Listwise: you want to delete any subject that has no value even in one variable (Will entirely
ignore participant 6)
Worth checking:

 The connection is not linear but there is a connection

 Are there outliers (extreme data)

Steps:

Graphs  Legacy Dialogs  Scatter

Dots  Simple Scatter  Define

Exceptional observation is an observation that is inconsistent with the vast majority of observations.

Sometimes we will observe outliers when there is a relevant and defined condition for this. Of course we
will report this in the results section

 There is no connection at all

 There is a nonlinear relationship
 There are unusual observations in a direction that is contrary to the general relationship

Example of write up:

A significant positive correlation was found between the psychometric score and the BA degree

(rp = 0.78, p <0.001). The significance of the finding is that the higher the psychometric score, the higher
the BA score. The strength of the received connection is high.

Spearman
  use when at least one of the two variables is ordinal and of course if both are Ordinal.
 A variable will be considered an ordinal if it has at least 3 values.
 done the same way as Pearson, but choose Spearman
 The test and its polynomials are almost identical to Pearson, except for the variable scale

Examples when to use:

 The number of visits per year, and the auditor's level of anxiety
 The number of delays to practical work, and the ranking of the instructor at the end of the year.
 Rating level of satisfaction with course, and rating level of satisfaction with grade in the course.

Overview:

we actually only learned two types of tests:

 T tests and Anova tests - compare averages

 Chai squared tests and correlations - test relationship between variables

effect size has an additional function: possibility of comparing effect sizes between different tests.
seeing if a result in one test (Anova) is more significant than a result in another test (chai-squared).
 Validity

when planning research the tool used should be validated

A major part of validity is reliability: Is the research tool I used reliable.

Cronbach's alpha

 Are the items in the tool testing the same content (internal consistency within tools)
 examines correlation between all items in the research tool.
 High alpha indicates pattern of responses to the various questions is consistent among subjects.
 Higher the correlation, the more reliable the test, - proves the questions relate to same content.
 The alpha range is between 0 and 1 (similar to correlations tests)

should be emphasized that the interest is the

profile of the answers, not the answers
themselves, so an inverted profile will also
promote the reliability of the items

accepted that alpha of 0.6-0.9 or higher indicates

reasonable internal consistency.

depending on the objective, we can settle for the

maximum value with the items we have. Example:
if omitting an item or two increases consistency
from 0.4 to 0.55, this is significant for
understanding the nature of the questions.
AnalyseScalereliability
analysisstatisticitem, scale, scale
if item deleted

scale if item deleted shows what

would happen if this question was
deleted from the test
 look at final column

How to write findings:

An inter-item reliability test was conducted to examine the internal consistency of the X research tools.
An internal consistency level of Alpha Kornbach of Y. was found. After we omitted Z1 and Z2, which
showed a weaker link to the content world of the research tool, the cornebach alpha increased to T,
Internal nicely between the N-2 questions remaining in the questionnaire.

We may also want to provide the reader with a list of questions or a list of questions that have been
omitted and a rationale for their lack of connection to the world of content.

Syntax

Paste: Save actions that have been made, reusable. To work with a lot data must learn to use this ability.

Do not forget to save the Syntax file!

Actions saved in Syntax allow for repeated analysis of the data at a different time, sending analytical
methods that you have produced to other researchers, and of course more accurate and rapid changes
of analyzes when needed, as well as creating new ones.

Things not covered in the course

 Two way ANOVA, MANOVA

 Repeated measures, Mixed model ANOVA
 Partial correlation
 Moderation and mediation (process – Hayes)
 Multiple linear regression/logistic regression
 Discriminant analysis
 Factor analysis, Cluster analysis
 MDS, SEM…