As is defined in :http://searchbusinessanalytics.techtarget.com/definition/association-rules-in-data-mining
An association rule has two parts, an antecedent (if) and a consequent (then). An antecedent is an item found in the data. A consequent is an item that is found in combination with the antecedent.
Association rules are created by analyzing data for frequent if/then patterns and using the criteria support and confidence to identify the most important relationships. Support is an indication of how frequently the items appear in the database. Confidence indicates the number of times the if/then statements have been found to be true.
In data mining, association rules are useful for analyzing and predicting customer behavior. They play an important part in shopping basket data analysis, product clustering, catalog design and store layout.
Could I use it as a rule (cause and effect) in the objective to validate it using Structual Equation model?
Certainly not. The IF ... THEN of an association rule represents a discovered correlation, not a cause and effect relationship. Consider the situation of mining shopping preferences and discovering that IF buys(cat_litter) THEN buys (cat_food). Does this allow you to infer that cat_litter purchases CAUSE cat_food purchases? No. If anything it would be the other way around (what goes in must come out) but you are unlikely to discover this reverse relationship anyway because not all cats, for whom cat_food is purchased, will use litter trays, for which cat_litter would be purchased.
.
an association rule is definitely not a proxy for a cause and effect relationship ; association rule is not even a very good proxy for correlation (you learn nothing on what happens outside the rule body)
- conducting a statistical analysis is about correlations (loosely speaking ...) between measurements
- building models is about tracing the causal relationships between events
- designing experiments is about linking said events to said measurements
.
the above is obviously much too rough but this is not the place to discuss the Humean induction problem !
.
now, the above does not mean that you cannot use association rule mining as a tool inside a structural equation model, although i am unsure of how this is done technically : a structural equation model contains the first two parts above (causal model + statistical modelling) !
.
.
Leaving aside the closed world assumption, SEMs should be used to test potential causal models for which there is some empirical or theoretical reason to posit them in the first place.
Thank you a lot, and I agree with all of you.
Could I interpret the following rule : the appearance of A and B imply the Appearance of C and D from a database ? during the process of extraction if there is no appearance of A and B cause the no appearance of C and D which constitutes the consequent of the rule. Cause and effect are implicit here, no?
I'm not too clear what you are asking here but I think you might be confusing a logical consequence with cause and effect. They are not the same. Also, if you have somehow determined that (A AND B) => (C AND D) then remember that a FALSE can imply anything in logic so you cannot infer anything from NOT (A AND B).
.
the answer is NO ... the "=>" symbol in the rule does not bear any causal meaning, it is strictly associational
If you want to dive into the details of the difference between "causal" and "associational", have a look a the Judea Pearl's paper Causal inference in statistics (attached below)
.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling