Need help with regression analysis

[tayvuihan]

Hi everyone, I am kinda confused with regression analysis. My research is to investigate the degree and trend women in the construction industry suffer from lower wage and harassment.

What I got here is a dependent variable (Salary) and independent variables
1. Gender
2. Years of working experience in the industry
3. Level of education
4. Working hours per week

What I am not sure of is what type of regression analysis technique to use. Linear regression/simple regression...etc



And I am also looking at

opportunity of promotion (Dependent Variable) by asking questions such as you felt that you need to work harder to gain promotion and salary received is in par with number of years of experience

Harassment (Dependent Variable) by asking questions such as you felt harassed in your workplace, you felt bullied in your workplace and you have seen a female colleague being harrassed

Psychological Environment (Dependent Variable) by asking questions such as you felt working with female colleague will generate better teamwork, you felt you receive less guidance than you colleague and proper facilities and amenities in your workplace

All by using scoring method (Strongly disagree to Strongly agree)

Other questions in the questionnaires are the age, annual salary, current working position and type of employer.

Is anyone able to point me to the right direction/advice me? I was told by my supervisor that Chi square is not appropriate to analyse my data and I will need to use regression analysis.

 

[Miner]

Multiple regression should work in this case. You do have to create indicator variables for some of your IVs (e.g., Gender, Educations etc.). You may also have to explore potential interactions and/or second order effects.

 

[Karabiner]

What I am not sure of is what type of regression analysis technique to use. Linear regression/simple regression...etc

Since salary is interval scaled, you can perform multiple linear regression
(maybe including gender x predictor interactions, if you are not only
interested in the direct effect of gender, but also in whether gender
moderates e.g. the effect of education on salary). There maybe will be
some technical difficulties when performing this, for example "level of
education" is an ordinal variable, which cannot easily be used as a predictor
in regression analyses. Also, prediction errors (residuals) from regression
models for salary data are often skewed, which could be a problem
if your sample isn't large.

Regarding the other dependent variables, they will probably be built by summing
the single items of the respective domains; I suppose they can be treated
as being interval scaled then, and therefore also be used as dependent variables
in multiple linear regression analyses.

With kind regards

K.

 

[rogojel]

hi,
for the likert scale DVs (I am assuming you use a Likert scale for the scoring) ordinal logistic regression could be an option.

regards
rogojel

 

[noetsi]

Education is probably best coded by a series of dummies depending on how you want to use it. For example if you think college is critical you could simply collapse the variable into college/no college and use one dummy. An alternative if you have years of education is simply to treat that as an interval variable although since most will probably have 12-18 years that is iffy in practice (not really enough distinct levels to be interval).

It is not uncommon in social science literature of use an ordinal variable with a few levels as if it was interval - but that creates I have been told probably with interpretation.