Correlation, ANOVA, t-test, chi-square have numeric dependent variables E.g. test score, number of words in corpus, F2, reaction Logistic Regression.

• Correlation, ANOVA, t-test, chi-square have numeric dependent variables• E.g. test score, number of words in corpus, F2, reaction

Logistic Regression

• Logistic regression has a categorical dependent variable• Expressed or implied pronoun

• I went and got vs. I went and I got

• Passive vs. active• John painted the barn vs. the barn was painted by John

• Deleted or pronounced• He wen’ home vs. he went home

Logistic Regression

• Logistic regression has a categorical dependent variable• Expressed or implied pronoun

• I went and got vs. I went and I got

• Passive vs. active• John painted the barn vs. the barn was painted by John

• Deleted or pronounced• He wen’ home vs. he went home

• The dependent variable can have 2 or more values• [t] vs. [ʔ] vs. []• Abou/t/ any

Logistic Regression

• Examples:

• Possession is expressed with both have, have got and got. What favors the use of each one? Could it be the animacy of the possessor? The animacy of the thing possessed? The socioeconomic status of the speaker? The frequency of the noun that follows?• Word-final /t/ and /d/ in words such as went and field is

sometimes pronounced and other times deleted. What is the role of the following consonant, the preceding vowel, the frequency of the word, the ethnicity of the speaker, and the age of the speaker in deletion or retention of these consonants?

Logistic Regression

• /t/ before /n/ in kitten, eaten, fountain• [khɪʔn]• [khɪʔən]• [khɪtən]

Logistic Regression


• The oral release is stigmatized in Utah

Logistic Regression


• The oral release is stigmatized in Utah

Logistic Regression

• What independent variables influence oral release versus the other two?• Gender• % of life in Utah• Age• Word order in passage read

Oral release in Utah

Example Study: Oral Release in Utah• Significance of the independent variables

•Doesn’t tell you direction of the effect

Example Study: Oral Release in Utah• Significance of the values of the independent

variables• Positive coefficients indicate a value favors oral releases

• Negative coefficients indicate a value disfavors oral releases

Example Study: Oral Release in Utah•Odds ratios (Exp(Coefficient) indicate the odds

increase.• The odds are over 11 times greater that women use oral

releases over men.

• For continuous variable like age, odds ratio of 1.053 means that for every percent increase in years living in Utah, the odds of using oral release increases 5.3%

Example Study: Oral Release in Utah

Results

•Odds ratios (Exp(Coefficient) indicate the odds increase.• Odds ratios under 1, such as .912 for age are subtracted

from 1

• 1-.912=.088

• The odds of using oral releases decreases by 8.8% for every year older the person is

Caveat: repeated measures• You can only include one measure/observation/score from a single person.• If you include more than one score you must include the participants as a special kind of variable (random effect)

Doing Mixed-effect Logistic Regression in SPSS

• In SPSS, click on Analyze > Mixed Models >Generalized Linear to open this box.

• All of the variables are listed here


• All of the variables are listed here

• The subject of the random effects, participant and word are dragged here


• In the Fields and Effects tab, Target is where you choose the dependent variable, Binary pronunciation


• In the Fields and Effects tab, Target is where you choose the dependent variable, Binary pronunciation

• ZOther is the value of the dependent variable we are going to contrast oral release to

• We are going to do Binary logistic regression


• By choosing Fixed Effects we see a list of variables to choose. Fixed effects are independent variables


• By choosing Fixed Effects we see a list of variables to choose

• We drag them into the next window


• Choosing Random Effects opens this window



• We choose Participant as the random intercept and leave the covariance type as variance component




• Be sure to check Include Intercept




• Be sure to check Include Intercept

• If you were doing a participant by other variable random slope you’d put the other variable here


• To add another random effect you click Add Block with brings up another window

• Choose Participant*Word > OK


• On the Build Options tab is the Sorting Order

• If you choose Ascending the value of the variable that occurs first alphabetically will be chosen as the reference value• Oral over Zother

• Descending would choose the last value alphabetically• Zother over Oral


• On the Build Options tab is the Sorting Order

• If you choose Ascending the value of the variable that occurs first alphabetically will be chosen as the reference value• Oral over Zother

• Descending would choose the last value alphabetically• Zother over Oral

• I already chose the value on the previous menu

• Press “paste”


• All of the code shows up.• We want a random effect

for participant and one for word so we have to tweak one line

• l/RANDOM USE_INTERCEPT=TRUE SUBJECTS=Participant*Word

• Delete “*Word”• Press green triangle to run

analysis.


• Double click on this table to open the Model Viewer window.

• Click on each sub-window on the left hand side to see the results.


• When you get this window choose Table under the Style dropdown menu on the bottom



• That gives you this table you’ve already seen.





• That gives you this table you’ve already seen.

• Click on the Coefficient and Exponential buttons to reveal the whole table.

Correlation, ANOVA, t-test, chi-square have numeric dependent variables E.g. test score, number of words in corpus, F2, reaction Logistic Regression.

Documents

oral release increases

odds increase

effect logistic regression

passage readoral release

homethe dependent variable

utahexample study

effectexample study

utahlogistic regressiont