SAS Tips - Comments

Comments are always import in any code. They make the code clearer, and are helpful in case of future consultations.  

To make this task easier here are some shortcuts:

• press [ctrl+/] to comment a line
• press [ctrl+shift+/] to uncomment a line

2013 International Year of Statistics

I would like to share 2013 the year of Statistics website:
http://www.statistics2013.org/

Take a look on how Statistics can contribute to many different problems across the world: http://www.statistics2013.org/files/2012/12/STAT2013Poster.pdf

Also, view the video https://www.youtube.com/watch?v=nTBZuQR7dRc, which relates the many ways that Statistics can impacts our lives.

Anyway, 2013 will come to an end (eventually), but Statistics rides on!

Survey entities

The figure below simplifies some of entities involved in a survey process. Can you tell which one is missing?

Basically a survey starts with the desire of a sponsor, who has a topic to cover.  This idea can have many uses, such as:

•                 analyze  process improvement or monitor its performance;
•                 implement a program evaluation;
•                 research about the quality of people’s work life;
•                 develop a market research regarding a specific product;
•                 investigate service customer satisfaction.

Once the topic is defined, the researcher will be response for the data collection and further analysis, providing final results to the sponsor.

The ideal survey

The ideal survey from the respondents’ perspective has:

• No costs
• High rewards

How to get there?

Help increase the response rate by reducing costs to respondents:

•                 Making the questions appear to be brief. For mail survey, one page is best.
•                 Reduce the physical/mental effort to answer the questions
•                 Eliminating any question that can cause embarrassment. Be aware that some questions can have a higher nonresponse rate such as family income, age sexual behavior etc. therefore avoid those. These questions can be interpreted be on sensitive or invasive topics.
•                 Respondent should never have any direct monetary cost

Help increase the response rate by rewarding the respondents:

•                 State at the beginning of the survey any tangible reward that will be offered.
•                 Make the questionnaire attention-grabbing
•                 Giving verbal appreciation and positive regard
•                 Establishing trust and using a consultative approach. Implication of subordination must be eliminated.

Starting Values - Nonlinear Models

Fitting a nonlinear regression requires starting values. The values used for the starting values are dependent on the model chosen. Of course, values close to the true parameter value will reduce convergence issues. A poor choice can cause convergence to a local minimum on the function, which can leads to a suboptimal solution. 

Monomolecular model:

Logistic Model:

Mitcherlich model:

Parameter Estimates - Nonlinear Models

In a simple nonlinear model, even with one parameter, the estimation is complicated. In order to remedy this, here there are the more common methods used for estimating parameters in nonlinear cases:

• ·         Linearization (Gauss-Newton) – Implemented in Minitab and SAS
• ·         Steepest Descent – Implemented in SAS (gradient)
• ·         Marquardt’s Compromise – Implemented in Minitab and SAS
• ·         Newton – Implemented in SAS

All techniques employ an iterative process until some minimum convergence has been reached. 

Types of nonlinear models

Growth models are typically mechanistic rather than empirical. Mechanistic models are developed based on some assumptions about the specific situation or environment. Based on these theoretical considerations equations are written. Thus, we do not need as much data to create a reliable model.

In empirical models, the mechanistic relationship is unknown, so we need a lot of data to show that the model is consistent.

In this post we discuss a few of these growth models based on theoretical considerations:

Monomolecular:

Properties:

When t → 0, then:     

When t → ∞, then the β term approaches 0 and W= α

There are no inflection points

Logistic:

Properties:

When t → 0, then: 

When t → ∞, then W= α

The curve is always positive.

Mitcherlich growth:

Properties:

Similar to the monomolecular growth it is used in chemical processes.